SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1612.00593 · CVPR · 2016

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

Kaichun Mo, Leonidas Guibas, Hao Su, Charles Qi

arXiv · PDF · Open in the Atlas

Code that ran

103 of 164 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (45 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 71 have no contract verdict; 32 failures of our instrument; 61 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/1612.00593.json · badge: /api/ran/1612.00593.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
ECE685-FinalProject/3D-object-recognition community
Papers with Code archive
7 ran · 2 no recorded run
SBPL-Cruz/perch_pose_sampler community (classifier label)
classifier basis not recorded on the link
7 ran · 0 no recorded run
SonuDileep/3-D-Object-Detection-using-PointNet community
not in the paper's declared repositories
6 ran · 6 no recorded run
PaParaZz1/PointNet community
Papers with Code archive
5 ran · 1 no recorded run
soumik12345/point-cloud-segmentation community
not in the paper's declared repositories
5 ran · 1 no recorded run
opeco17/pointnet community
not in the paper's declared repositories
5 ran · 0 no recorded run
KiranAkadas/My_Pointnet_v2 community
Papers with Code archive
4 ran · 8 no recorded run
y2kmz/pointnetv2 community
not in the paper's declared repositories
4 ran · 4 no recorded run
debajyotid2/pointnet_pl community
not in the paper's declared repositories
4 ran · 0 no recorded run
Taeuk-Jang/pointcompletion community
not in the paper's declared repositories
4 ran · 0 no recorded run
smfirmin/DeepLearningFinalProject community
not in the paper's declared repositories
3 ran · 3 no recorded run
minhncedutw/pointnet1_keras community
not in the paper's declared repositories
3 ran · 2 no recorded run
mengxingshifen1218/pointnet.pytorch community
not in the paper's declared repositories
3 ran · 1 no recorded run
romaintha/pytorch_pointnet community
not in the paper's declared repositories
3 ran · 0 no recorded run
brendanmallery9/entropic-barycenters community
not in the paper's declared repositories
3 ran · 0 no recorded run
nikitakaraevv/pointnet community
not in the paper's declared repositories
3 ran · 0 no recorded run
lingzhang1/pointnet_pytorch community
not in the paper's declared repositories
3 ran · 0 no recorded run
CheesyB/cpointnet community
Papers with Code archive
3 ran · 0 no recorded run
YanWei123/PointNet-encoder-and-FoldingNet-decoder-add-quantization-change-latent-code-size-from-512-to-1024 community
not in the paper's declared repositories
3 ran · 0 no recorded run
hnmspirit/pointnet1 community
not in the paper's declared repositories
3 ran · 0 no recorded run
yangyucheng000/pointnet community (classifier label)
classifier basis not recorded on the link
3 ran · 0 no recorded run
YanWei123/Pytorch-implementation-of-FoldingNet-encoder-and-decoder-with-graph-pooling-covariance-add-quanti community (classifier label)
classifier basis not recorded on the link
2 ran · 1 no recorded run
YanWei123/Pointnet_encoder_Foldingnet_decoder_quantization community (classifier label)
classifier basis not recorded on the link
2 ran · 0 no recorded run
vinits5/learning3d community
not in the paper's declared repositories
2 ran · 0 no recorded run
ftdlyc/pointnet_pytorch community
not in the paper's declared repositories
2 ran · 0 no recorded run
FunctionStatusWhere it lives
AttnEncoderBlock Ran
ran on a synthesized input
smfirmin/DeepLearningFinalProject/pointnet/model.py
code served (permissive licence) · get_code("89dc188151d58f24")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 64, 2500) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

class AttnEncoderBlock(nn.Module):

    def __init__(self, device, embed_dim=64, num_heads=2, norm='batch1d', dim_ff=128):
        super(AttnEncoderBlock, self).__init__()
        self.device = device
        self.num_heads = num_heads
        self.embed_dim = embed_dim
        self.dim_ff = dim_ff
        self.attn = nn.MultiheadAttention(embed_dim=self.embed_dim, num_heads=self.num_heads, batch_first=True)
        self.ff1 = nn.Linear(64, self.dim_ff)
        self.ff2 = nn.Linear(self.dim_ff, 64)
        self.relu = nn.ReLU()
        if norm == 'batch1d':
            self.norm1 = nn.BatchNorm1d(2500)
            self.norm2 = nn.BatchNorm1d(2500)
        elif norm == 'layer':
            self.norm1 = nn.LayerNorm(64)
            self.norm2 = nn.LayerNorm(64)

    def forward(self, x):
        x = x.transpose(2, 1)
        sa = self.attn(x, x, x)[0]
        x = self.norm1(x + sa)
        x = self.norm2(x + self.ff2(self.relu(self.ff1(x))))
        x = x.transpose(2, 1)
        return x


INIT_ARGS = {
    'device': 'cpu',
    'embed_dim': 64,
    'num_heads': 2,
    'norm': 'layer',
    'dim_ff': 128
}

ARGS = {
    'x': torch.randn(2, 64, 2500)
}

if __name__ == '__main__':
    model = AttnEncoderBlock(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: {ARGS['x'].shape}")
    assert output.shape == ARGS['x'].shape, f"Shape mismatch: {output.shape} vs {ARGS['x'].shape}"
    print("Success!")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 64, 2500) — shape (2, 64, 2500) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
BasePointNet Ran
ran on a synthesized input · constructed an object; no result was computed
romaintha/pytorch_pointnet/model/pointnet.py
code served (permissive licence) · get_code("6a7f90648d711cdc")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, BasePointNet: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "point_dimension": 3,
    "return_local_features": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type BasePointNet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a BasePointNet output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ChamferDistance Ran
ran on a synthesized input
YanWei123/Pointnet_encoder_Foldingnet_decoder_quantization/pointnet.py
code served (permissive licence) · get_code("2bb949769eebab26")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create two point clouds with compatible dimensions
# x and y must have:
# - Same batch size (dimension 0)
# - Same feature dimension (dimension 2)
# - Different point counts (dimensions 1)

batch_size = 2
num_points_x = 4
num_points_y = 3
feature_dim = 5

x = torch.randn(batch_size, num_points_x, feature_dim)
y = torch.randn(batch_size, num_points_y, feature_dim)

ARGS = {
    "x": x,
    "y": y,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ClassificationPointNet Ran
ran on a synthesized input · constructed an object; no result was computed
romaintha/pytorch_pointnet/model/pointnet.py
code served (permissive licence) · get_code("0ba84a38091fc6f1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, ClassificationPointNet: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'num_classes': 10,
    'dropout': 0.3,
    'point_dimension': 3,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ClassificationPointNet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ClassificationPointNet output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ConvBlock Ran
ran on a synthesized input · constructed an object; no result was computed
debajyotid2/pointnet_pl/src/model.py
pointer only (licence: NONE) · get_code("b5cf4f63f0bd5977")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, debajyotid2/pointnet_pl when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ConvBlock Ran
ran; our contract draft was wrong, not the code
PaParaZz1/PointNet/network.py
code served (permissive licence) · get_code("0500691c048885c9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'in_channels': 3,
    'out_channels': 16,
    'kernel_size': 5,
    'stride': 1,
    'padding': 2,
    'init_type': 'kaiming',
    'activation': nn.ReLU(),
    'use_batchnorm': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ConvBlockSequential Ran
ran; our contract draft was wrong, not the code
PaParaZz1/PointNet/network.py
code served (permissive licence) · get_code("d35a1145885d3159")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Sequential. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.nn.init import xavier_normal, kaiming_normal

torch.manual_seed(0)

ARGS = {
    'in_channels': 3,
    'out_channels': 16,
    'kernel_size': 5,
    'stride': 1,
    'padding': 2,
    'init_type': 'kaiming',
    'activation': nn.ReLU(),
    'use_batchnorm': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential output the recorded ndim 1 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
DensityNet Ran
ran on a synthesized input · constructed an object; no result was computed
ECE685-FinalProject/3D-object-recognition/model/pointconv.py
pointer only (licence: NONE) · get_code("101276f7d87be7cb")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ECE685-FinalProject/3D-object-recognition when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

FCBlock Ran
ran on a synthesized input · constructed an object; no result was computed
debajyotid2/pointnet_pl/src/model.py
pointer only (licence: NONE) · get_code("c0998ae0e66ae44b")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, debajyotid2/pointnet_pl when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

FeatureTNet Ran
ran on a synthesized input · constructed an object; no result was computed
opeco17/pointnet/model.py
pointer only (licence: NONE) · get_code("f153c54ecb32c4eb")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, opeco17/pointnet when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

FeatureTransform Ran
ran on a synthesized input
PaParaZz1/PointNet/network.py
code served (permissive licence) · get_code("e7a88d785b2894ba")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 4, 4) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
from torch.nn.init import xavier_normal, kaiming_normal

def ConvBlock(in_channels, out_channels, kernel_size, stride=1, padding=0, init_type='kaiming', activation=nn.ReLU(), use_batchnorm=False):
    block_conv = []
    block_conv.append(nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding))
    WeightInitialize(init_type, block_conv[-1].weight, activation)
    if not activation is None:
        block_conv.append(activation)
    if use_batchnorm:
        block_conv.append(nn.BatchNorm1d(out_channels))
    return block_conv

def ConvBlockSequential(in_channels, out_channels, kernel_size, stride=1, padding=0, init_type='kaiming', activation=nn.ReLU(), use_batchnorm=False):
    seq = nn.Sequential(*ConvBlock(in_channels, out_channels, kernel_size, stride, padding, init_type, activation, use_batchnorm))
    seq.out_channels = out_channels
    return seq

def WeightInitialize(init_type, weight, activation=None):
    if init_type is None:
        return

    def XavierInit(weight, activation):
        xavier_normal(weight)

    def KaimingInit(weight, activation):
        assert not activation is None
        if hasattr(activation, 'negative_slope'):
            kaiming_normal(weight, a=activation.negative_slope)
        else:
            kaiming_normal(weight, a=0)
    init_type_dict = {'xavier': XavierInit, 'kaiming': KaimingInit}
    if init_type in init_type_dict:
        init_type_dict[init_type](weight, activation)
    else:
        raise KeyError('Invalid Key:%s' % init_type)

class FeatureTransform(nn.Module):

    def __init__(self, k=64):
        super(FeatureTransform, self).__init__()
        self.k = k
        self.conv1 = ConvBlockSequential(in_channels=64, out_channels=128, kernel_size=1, init_type='xavier', use_batchnorm=True)
        self.conv2 = ConvBlockSequential(in_channels=128, out_channels=256, kernel_size=1, init_type='xavier', use_batchnorm=True)
        self.conv3 = ConvBlockSequential(in_channels=256, out_channels=1024, kernel_size=1, init_type='xavier', use_batchnorm=True)
        self.fc1 = nn.Linear(1024, 2048)
        self.fc2 = nn.Linear(2048, k * k)
        self.act = nn.PReLU(init=0.1)
        self.dropout1 = nn.Dropout(0.2)
        self.dropout2 = nn.Dropout(0.3)

    def forward(self, x):
        batch_size = x.size()[0]
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        x = torch.max(x, 2)[0]
        x = x.view(-1, 1024)
        x = self.dropout1(x)
        x = self.act(self.fc1(x))
        x = self.dropout2(x)
        x = self.fc2(x)
        iden = Variable(torch.from_numpy(np.eye(self.k).astype(np.float32))).view(1, self.k * self.k).repeat(batch_size, 1)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, self.k, self.k)
        return x

INIT_ARGS = {
    'k': 4
}

ARGS = {
    'x': torch.randn(2, 64, 3)
}

if __name__ == '__main__':
    model = FeatureTransform(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 4, 4)")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 4, 4) — shape (2, 4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
GridSamplingLayer Ran
ran; our contract draft was wrong, not the code
YanWei123/Pointnet_encoder_Foldingnet_decoder_quantization/pointnet.py
code served (permissive licence) · get_code("f0f053f931f94251")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family float, shape (2, 20, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

ARGS = {
    'batch_size': 2,
    'meshgrid': [[-0.3, 0.3, 5], [-0.5, 0.5, 4]]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (2, 20, 2) — shape (2, 20, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
InputTNet Ran
ran on a synthesized input · constructed an object; no result was computed
opeco17/pointnet/model.py
pointer only (licence: NONE) · get_code("f7820c2d16a9147f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, opeco17/pointnet when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

InputTransform Ran
ran on a synthesized input
PaParaZz1/PointNet/network.py
code served (permissive licence) · get_code("d4c6fdd6cac8cdfa")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
from torch.nn.init import xavier_normal, kaiming_normal

def ConvBlock(in_channels, out_channels, kernel_size, stride=1, padding=0, init_type='kaiming', activation=nn.ReLU(), use_batchnorm=False):
    block_conv = []
    block_conv.append(nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding))
    WeightInitialize(init_type, block_conv[-1].weight, activation)
    if not activation is None:
        block_conv.append(activation)
    if use_batchnorm:
        block_conv.append(nn.BatchNorm1d(out_channels))
    return block_conv

def ConvBlockSequential(in_channels, out_channels, kernel_size, stride=1, padding=0, init_type='kaiming', activation=nn.ReLU(), use_batchnorm=False):
    seq = nn.Sequential(*ConvBlock(in_channels, out_channels, kernel_size, stride, padding, init_type, activation, use_batchnorm))
    seq.out_channels = out_channels
    return seq

def WeightInitialize(init_type, weight, activation=None):
    if init_type is None:
        return

    def XavierInit(weight, activation):
        xavier_normal(weight)

    def KaimingInit(weight, activation):
        assert not activation is None
        if hasattr(activation, 'negative_slope'):
            kaiming_normal(weight, a=activation.negative_slope)
        else:
            kaiming_normal(weight, a=0)
    init_type_dict = {'xavier': XavierInit, 'kaiming': KaimingInit}
    if init_type in init_type_dict:
        init_type_dict[init_type](weight, activation)
    else:
        raise KeyError('Invalid Key:%s' % init_type)

class InputTransform(nn.Module):

    def __init__(self, k=4):
        super(InputTransform, self).__init__()
        self.k = k
        self.conv1 = ConvBlockSequential(in_channels=4, out_channels=64, kernel_size=1, init_type='xavier', use_batchnorm=True)
        self.conv2 = ConvBlockSequential(in_channels=64, out_channels=128, kernel_size=1, init_type='xavier', use_batchnorm=True)
        self.conv3 = ConvBlockSequential(in_channels=128, out_channels=512, kernel_size=1, init_type='xavier', use_batchnorm=True)
        self.conv4 = ConvBlockSequential(in_channels=512, out_channels=1024, kernel_size=1, init_type='xavier', use_batchnorm=True)
        self.fc1 = nn.Linear(1024, 512)
        self.bn1 = nn.BatchNorm2d(512)
        self.fc2 = nn.Linear(512, 256)
        self.bn2 = nn.BatchNorm2d(256)
        self.fc3 = nn.Linear(256, 4 * k)
        self.act = nn.PReLU(init=0.1)
        self.dropout = nn.Dropout(0.1)

    def forward(self, x):
        batch_size = x.size()[0]
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)
        x = torch.max(x, 2)[0]
        x = x.view(-1, 1024)
        x = self.dropout(x)
        x = self.act(self.fc1(x))
        x = self.act(self.fc2(x))
        x = self.fc3(x)
        iden = Variable(torch.from_numpy(np.eye(self.k).astype(np.float32))).view(1, self.k * 4).repeat(batch_size, 1)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, 4, self.k)
        return x

INIT_ARGS = {
    'k': 4
}

ARGS = {
    'x': torch.randn(2, 4, 128)
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MatMul Ran
ran on a synthesized input
AI-Guru/pointcloud_experiments/model_cls.py
pointer only (licence: NONE) · get_code("3829b0b4b30416c1")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, AI-Guru/pointcloud_experiments when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MaxPool Ran
ran on a synthesized input · constructed an object; no result was computed
opeco17/pointnet/model.py
pointer only (licence: NONE) · get_code("0bcd28a18c4f1eac")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, opeco17/pointnet when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

NonLinear Ran
ran on a synthesized input · constructed an object; no result was computed
opeco17/pointnet/model.py
pointer only (licence: NONE) · get_code("44f26f82c86e21d7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, opeco17/pointnet when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

OrthogonalRegularizer Ran
ran on a synthesized input · constructed an object; no result was computed
soumik12345/point-cloud-segmentation/point_seg/pointnet/models.py
code served (permissive licence) · get_code("63fc528b5460f4b1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, OrthogonalRegularizer: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tensorflow as tf
from tensorflow.keras import layers, regularizers, initializers

# Set random seeds for reproducibility
tf.random.set_seed(0)

ARGS = {
    'num_features': 4,
    'l2reg': 0.001
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type OrthogonalRegularizer — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a OrthogonalRegularizer output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNet Ran
ran on a synthesized input · constructed an object; no result was computed
vinits5/learning3d/models/pointnet.py
code served (permissive licence) · get_code("90de9a977caa457e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNet: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    'emb_dims': 1024,
    'input_shape': 'bnc',
    'use_bn': False,
    'global_feat': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNet output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNet Ran
ran on a synthesized input · constructed an object; no result was computed
WangYueFt/dgcnn/pytorch/model.py
code served (permissive licence) · get_code("d49fb29dc696ee6c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNet: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class ArgsConfig:
    def __init__(self, emb_dims=1024):
        self.emb_dims = emb_dims

ARGS = {
    'args': ArgsConfig(emb_dims=1024),
    'output_channels': 40
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNet output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNet Ran
ran on a synthesized input · constructed an object; no result was computed
brendanmallery9/entropic-barycenters/classification_example/pointnet_implementation.py
code served (permissive licence) · get_code("df9cbdf65f14db23")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNet: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "classes": 5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNet output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNet Ran
ran on a synthesized input · constructed an object; no result was computed
opeco17/pointnet/model.py
pointer only (licence: NONE) · get_code("f926ecb7397001aa")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, opeco17/pointnet when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PointNet Ran
ran on a synthesized input
nikitakaraevv/pointnet/source/model.py
code served (permissive licence) · get_code("0bb3296226c3f2e6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned tuple. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

# PointNet class definition (from the provided code)
class Tnet(nn.Module):

    def __init__(self, k=3):
        super().__init__()
        self.k = k
        self.conv1 = nn.Conv1d(k, 64, 1)
        self.conv2 = nn.Conv1d(64, 128, 1)
        self.conv3 = nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k * k)
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, input):
        bs = input.size(0)
        xb = F.relu(self.bn1(self.conv1(input)))
        xb = F.relu(self.bn2(self.conv2(xb)))
        xb = F.relu(self.bn3(self.conv3(xb)))
        pool = nn.MaxPool1d(xb.size(-1))(xb)
        flat = nn.Flatten(1)(pool)
        xb = F.relu(self.bn4(self.fc1(flat)))
        xb = F.relu(self.bn5(self.fc2(xb)))
        init = torch.eye(self.k, requires_grad=True).repeat(bs, 1, 1)
        if xb.is_cuda:
            init = init.cuda()
        matrix = self.fc3(xb).view(-1, self.k, self.k) + init
        return matrix

class Transform(nn.Module):

    def __init__(self):
        super().__init__()
        self.input_transform = Tnet(k=3)
        self.feature_transform = Tnet(k=64)
        self.conv1 = nn.Conv1d(3, 64, 1)
        self.conv2 = nn.Conv1d(64, 128, 1)
        self.conv3 = nn.Conv1d(128, 1024, 1)
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)

    def forward(self, input):
        matrix3x3 = self.input_transform(input)
        xb = torch.bmm(torch.transpose(input, 1, 2), matrix3x3).transpose(1, 2)
        xb = F.relu(self.bn1(self.conv1(xb)))
        matrix64x64 = self.feature_transform(xb)
        xb = torch.bmm(torch.transpose(xb, 1, 2), matrix64x64).transpose(1, 2)
        xb = F.relu(self.bn2(self.conv2(xb)))
        xb = self.bn3(self.conv3(xb))
        xb = nn.MaxPool1d(xb.size(-1))(xb)
        output = nn.Flatten(1)(xb)
        return (output, matrix3x3, matrix64x64)

class PointNet(nn.Module):

    def __init__(self, classes=10):
        super().__init__()
        self.transform = Transform()
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, classes)
        self.bn1 = nn.BatchNorm1d(512)
        self.bn2 = nn.BatchNorm1d(256)
        self.dropout = nn.Dropout(p=0.3)
        self.logsoftmax = nn.LogSoftmax(dim=1)

    def forward(self, input):
        xb, matrix3x3, matrix64x64 = self.transform(input)
        xb = F.relu(self.bn1(self.fc1(xb)))
        xb = F.relu(self.bn2(self.dropout(self.fc2(xb))))
        output = self.fc3(xb)
        return (self.logsoftmax(output), matrix3x3, matrix64x64)


# Setup code
INIT_ARGS = {
    'classes': 5
}

ARGS = {
    'input': torch.randn(2, 3, 8)
}

# Construct and call
model = PointNet(**INIT_ARGS)
model.eval()
output = model(**ARGS)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
PointNetClassifier Ran
ran on a synthesized input · constructed an object; no result was computed
debajyotid2/pointnet_pl/src/model.py
pointer only (licence: NONE) · get_code("86c240d726febafb")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, debajyotid2/pointnet_pl when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PointNetClassify Ran
ran on a synthesized input · constructed an object; no result was computed
ftdlyc/pointnet_pytorch/models/pointnet.py
pointer only (licence: NONE) · get_code("05f62796540f4175")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ftdlyc/pointnet_pytorch when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PointNetCls Ran
ran on a synthesized input · constructed an object; no result was computed
lingzhang1/pointnet_pytorch/pointnet.py
code served (permissive licence) · get_code("21086233dee4a940")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetCls: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a minimal valid input tensor
# The model expects input shape: (batch_size, 6, num_points)
# where 6 channels come from concatenating 3D coordinates with something else
batch_size = 2
num_points = 2500
num_classes = 2

# Create input: (batch_size=2, channels=6, num_points=2500)
x = torch.randn(batch_size, 6, num_points)

ARGS = {
    "num_points": num_points,
    "k": num_classes,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetCls — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetCls output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetCls Ran
ran on a synthesized input · constructed an object; no result was computed
CheesyB/cpointnet/pointnet.py
code served (permissive licence) · get_code("26589a98563a59d0")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetCls: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
import torch.nn.functional as F

torch.manual_seed(0)

class STN3d(nn.Module):

    def __init__(self, num_points=2500):
        super(STN3d, self).__init__()
        self.num_points = num_points
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.mp1 = torch.nn.MaxPool1d(num_points)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, 9)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, x):
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = self.mp1(x)
        x = x.view(-1, 1024)
        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)
        iden = Variable(torch.from_numpy(np.array([1, 0, 0, 0, 1, 0, 0, 0, 1]).astype(np.float32))).view(1, 9).repeat(batchsize, 1)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, 3, 3)
        return x

class PointNetfeat(nn.Module):

    def __init__(self, num_points=2500, global_feat=True):
        super(PointNetfeat, self).__init__()
        self.stn = STN3d(num_points=num_points)
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.mp1 = torch.nn.MaxPool1d(num_points)
        self.num_points = num_points
        self.global_feat = global_feat

    def forward(self, x):
        batchsize = x.size()[0]
        trans = self.stn(x)
        x = x.transpose(2, 1)
        x = torch.bmm(x, trans)
        x = x.transpose(2, 1)
        x = F.relu(self.bn1(self.conv1(x)))
        pointfeat = x
        x = F.relu(self.bn2(self.conv2(x)))
        x = self.bn3(self.conv3(x))
        x = self.mp1(x)
        x = x.view(-1, 1024)
        if self.global_feat:
            return (x, trans)
        else:
            x = x.view(-1, 1024, 1).repeat(1, 1, self.num_points)
            return (torch.cat([x, pointfeat], 1), trans)

class PointNetCls(nn.Module):

    def __init__(self, num_points=2500, k=2):
        super(PointNetCls, self).__init__()
        self.num_points = num_points
        self.feat = PointNetfeat(num_points, global_feat=True)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k)
        self.bn1 = nn.BatchNorm1d(512)
        self.bn2 = nn.BatchNorm1d(256)
        self.relu = nn.ReLU()

    def forward(self, x):
        x, trans = self.feat(x)
        x = F.relu(self.bn1(self.fc1(x)))
        x = F.relu(self.bn2(self.fc2(x)))
        x = self.fc3(x)
        return (F.log_softmax(x, dim=-1), trans)

ARGS = {
    "num_points": 4,
    "k": 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetCls — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetCls output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetCls Ran
ran on a synthesized input · constructed an object; no result was computed
YanWei123/PointNet-encoder-and-FoldingNet-decoder-add-quantization-change-latent-code-size-from-512-to-1024/pointnet.py
code served (permissive licence) · get_code("6dcbc51747a8bb73")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetCls: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import numpy as np
from torch.autograd import Variable

torch.manual_seed(0)

# Define the STN3d and PointNetfeat classes as provided
class STN3d(nn.Module):
    def __init__(self):
        super(STN3d, self).__init__()
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, 9)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, x):
        import torch.nn.functional as F
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)
        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)
        iden = Variable(torch.from_numpy(np.array([1, 0, 0, 0, 1, 0, 0, 0, 1]).astype(np.float32))).view(1, 9).repeat(batchsize, 1)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, 3, 3)
        return x

class PointNetfeat(nn.Module):
    def __init__(self, global_feat=True):
        super(PointNetfeat, self).__init__()
        self.stn = STN3d()
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.global_feat = global_feat

    def forward(self, x):
        import torch.nn.functional as F
        batchsize = x.size()[0]
        n_pts = x.size()[2]
        trans = self.stn(x)
        x = x.transpose(2, 1)
        x = torch.bmm(x, trans)
        x = x.transpose(2, 1)
        x = F.relu(self.bn1(self.conv1(x)))
        pointfeat = x
        x = F.relu(self.bn2(self.conv2(x)))
        x = self.bn3(self.conv3(x))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)
        if self.global_feat:
            return (x, trans)
        else:
            x = x.view(-1, 1024, 1).repeat(1, 1, n_pts)
            return (torch.cat([x, pointfeat], 1), trans)

class PointNetCls(nn.Module):
    def __init__(self, k=2):
        super(PointNetCls, self).__init__()
        self.feat = PointNetfeat(global_feat=True)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k)
        self.bn1 = nn.BatchNorm1d(512)
        self.bn2 = nn.BatchNorm1d(256)
        self.relu = nn.ReLU()

    def forward(self, x):
        import torch.nn.functional as F
        x, trans = self.feat(x)
        x = F.relu(self.bn1(self.fc1(x)))
        x = F.relu(self.bn2(self.fc2(x)))
        x = self.fc3(x)
        return (F.log_softmax(x, dim=0), trans)

# Create a minimal input tensor: batch_size=2, channels=3, num_points=4
input_tensor = torch.randn(2, 3, 4)

ARGS = {
    'k': 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetCls — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetCls output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetCls Ran
ran on a synthesized input
hnmspirit/pointnet1/pointnet.py
pointer only (licence: NONE) · get_code("98f3e76878c4c84f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, hnmspirit/pointnet1 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PointNetFeat Ran
ran on a synthesized input
hnmspirit/pointnet1/pointnet.py
pointer only (licence: NONE) · get_code("5f85b825ddcf36f0")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, hnmspirit/pointnet1 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PointNetMLP Ran
ran on a synthesized input · constructed an object; no result was computed
ali-stanford/pointnetcfd/PyTorchVersion/PointNetCFD.py
code served (permissive licence) · get_code("0bf545f1ebcb90c5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetMLP: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# PointNetMLP expects:
# - input_channels: int (number of input channels for Conv1d)
# - output_channels: int (number of output channels for Conv1d)
# - scaling: float (optional, defaults to 1.0)

ARGS = {
    'input_channels': 3,
    'output_channels': 2,
    'scaling': 1.0
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetMLP — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetMLP output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetSeg Ran
ran on a synthesized input
PaParaZz1/PointNet/network.py
code served (permissive licence) · get_code("8727e973f8a99293")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned tuple. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

INIT_ARGS = {
    'num_class': 8
}

# Create input tensor: batch_size=2, sequence_length=5, feature_dim=4
input_tensor = torch.randn(2, 4, 5)

ARGS = {
    'x': input_tensor
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
mengxingshifen1218/pointnet.pytorch/pointnet/model.py
code served (permissive licence) · get_code("354def71e0aa90c3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetfeat: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import numpy as np
import logging

torch.manual_seed(0)
logging.basicConfig(level=logging.INFO)

# The PointNetfeat.__init__ takes two parameters:
# - global_feat (default True)
# - feature_transform (default False)

ARGS = {
    'global_feat': True,
    'feature_transform': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetfeat — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetfeat output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
lingzhang1/pointnet_pytorch/pointnet.py
code served (permissive licence) · get_code("4974babafbd20779")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetfeat: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# PointNetfeat expects:
# - num_points: int (default 2500)
# - global_feat: bool (default True)

# Looking at the forward method:
# - x is transposed and converted to numpy, then used in bmm
# - x should have shape (batch, num_points, features) based on transpose(2,1) operations
# - The stn processes x and returns trans of shape (batch, 6, 6)
# - bmm operation: x (batch, num_points, 6) @ trans (batch, 6, 6) -> (batch, num_points, 6)

# Create a small input tensor
# batch_size = 2, num_points = 4, num_features = 6
batch_size = 2
num_points = 4
num_features = 6

# Input tensor: (batch, num_features, num_points) for Conv1d
x = torch.randn(batch_size, num_features, num_points)

ARGS = {
    "num_points": num_points,
    "global_feat": True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetfeat — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetfeat output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
yanxp/PointNet/model/pointnet.py
pointer only (licence: NONE) · get_code("674ede50cbdb04c2")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, yanxp/PointNet when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
CheesyB/cpointnet/pointnet.py
code served (permissive licence) · get_code("62f355b393ae1ca7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetfeat: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import numpy as np

torch.manual_seed(0)

ARGS = {
    "num_points": 4,
    "global_feat": True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetfeat — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetfeat output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
YanWei123/PointNet-encoder-and-FoldingNet-decoder-add-quantization-change-latent-code-size-from-512-to-1024/pointnet.py
code served (permissive licence) · get_code("5a5a64fad512c4b2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetfeat: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import numpy as np

torch.manual_seed(0)

ARGS = {
    'global_feat': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetfeat — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetfeat output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
Taeuk-Jang/pointcompletion/pointnet/model.py
code served (permissive licence) · get_code("a0a98d1455e2700c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetfeat: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import numpy as np

torch.manual_seed(0)

# Device should be CPU only
device = 'cpu'

# Create minimal input tensor
# x should have shape (batch_size, channels, num_points)
# For channel=4, we need 4 input channels
x = torch.randn(2, 4, 8)

ARGS = {
    'device': device,
    'channel': 4,
    'global_feat': True,
    'feature_transform': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetfeat — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetfeat output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
princeton-vl/SimpleView/pointnet_pyt/pointnet/model.py
code served (permissive licence) · get_code("3ce362b43484ed47")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetfeat: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import numpy as np

torch.manual_seed(0)

# Create a small point cloud batch
# Shape: (batch_size, num_features, num_points)
# batch_size=2, num_features=3 (x,y,z coordinates), num_points=4
point_cloud = torch.randn(2, 3, 4)

ARGS = {
    "global_feat": True,
    "feature_transform": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetfeat — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetfeat output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
sarthakTUM/roofn3d/cls_seg_mt/pointnet.py
pointer only (licence: NONE) · get_code("79e1c88294831c30")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, sarthakTUM/roofn3d when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PointNetfeat Ran
ran on a synthesized input · constructed an object; no result was computed
xurui1217/pointnet.pytorch-master/pointnet.py
code served (permissive licence) · get_code("65e190258615dca9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PointNetfeat: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import numpy as np

torch.manual_seed(0)

ARGS = {
    'global_feat': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PointNetfeat — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PointNetfeat output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Pooling Ran
ran on a synthesized input · constructed an object; no result was computed
vinits5/learning3d/models/pointnet.py
code served (permissive licence) · get_code("73fbe9c91fe32ceb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Pooling: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# The Pooling module's forward method expects:
# - input: a tensor that supports torch.max and torch.mean along dimension 2
# - The output is indexed with [0] for max, suggesting a 3D tensor
# - Contiguous() is called on the result

# Create a small 3D input tensor: (batch_size, feature_dim, sequence_length)
# Using distinct sizes: batch=2, features=4, sequence=3
input_tensor = torch.randn(2, 4, 3)

ARGS = {
    'pool_type': 'max'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Pooling — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Pooling output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STN3d Ran
ran on a synthesized input
mengxingshifen1218/pointnet.pytorch/pointnet/model.py
code served (permissive licence) · get_code("c44477c4a9ec175d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import logging
import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
import torch.nn.functional as F

logging.basicConfig(level=logging.INFO)

class STN3d(nn.Module):

    def __init__(self):
        super(STN3d, self).__init__()
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, 9)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, x):
        batchsize = x.size()[0]
        logging.info('The x type is:%s,shape is:%s', str(type(x)), str(x.shape))
        logging.info('The batchsize type is:%s,num is:%d', str(type(batchsize)), batchsize)
        x = F.relu(self.bn1(self.conv1(x)))
        logging.info('The x1 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = F.relu(self.bn2(self.conv2(x)))
        logging.info('The x2 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = F.relu(self.bn3(self.conv3(x)))
        logging.info('The x3 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = torch.max(x, 2, keepdim=True)[0]
        logging.info('The x4 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = x.view(-1, 1024)
        logging.info('The x5 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = F.relu(self.bn4(self.fc1(x)))
        logging.info('The x6 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = F.relu(self.bn5(self.fc2(x)))
        logging.info('The x7 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = self.fc3(x)
        logging.info('The x8 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        iden = Variable(torch.from_numpy(np.array([1, 0, 0, 0, 1, 0, 0, 0, 1]).astype(np.float32))).view(1, 9).repeat(batchsize, 1)
        logging.info('The iden type is:%s,shape is:%s', str(type(iden)), str(iden.shape))
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, 3, 3)
        logging.info('The x9 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        return x


INIT_ARGS = {}

ARGS = {
    'x': torch.randn(2, 3, 16)
}

if __name__ == '__main__':
    model = STN3d(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shape:", output.shape)
    print("Output:", output)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
STN3d Ran
ran on a synthesized input · constructed an object; no result was computed
lingzhang1/pointnet_pytorch/pointnet.py
code served (permissive licence) · get_code("83226be05a2733e2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, STN3d: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "num_points": 2500,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type STN3d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a STN3d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STN3d Ran
ran on a synthesized input · constructed an object; no result was computed
CheesyB/cpointnet/pointnet.py
code served (permissive licence) · get_code("4626eef340141c25")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, STN3d: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np

torch.manual_seed(0)

# The STN3d class expects:
# - num_points: int parameter used in MaxPool1d and to repeat the identity matrix

# Create a small input tensor for testing
# Shape: (batch_size, 3, num_points)
# Using batch_size=2, num_points=2500 as default
batch_size = 2
num_points = 2500
input_tensor = torch.randn(batch_size, 3, num_points)

ARGS = {
    "num_points": num_points,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type STN3d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a STN3d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STN3d Ran
ran on a synthesized input · constructed an object; no result was computed
YanWei123/PointNet-encoder-and-FoldingNet-decoder-add-quantization-change-latent-code-size-from-512-to-1024/pointnet.py
code served (permissive licence) · get_code("00b768e2c62c92bc")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, STN3d: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
import torch.nn.functional as F

torch.manual_seed(0)

class STN3d(nn.Module):

    def __init__(self):
        super(STN3d, self).__init__()
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, 9)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, x):
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)
        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)
        iden = Variable(torch.from_numpy(np.array([1, 0, 0, 0, 1, 0, 0, 0, 1]).astype(np.float32))).view(1, 9).repeat(batchsize, 1)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, 3, 3)
        return x

# Create input tensor with batch size 2, 3 channels, 4 points
input_tensor = torch.randn(2, 3, 4)

# STN3d.__init__ takes no parameters besides self
ARGS = {
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type STN3d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a STN3d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STN3d Ran
ran on a synthesized input · constructed an object; no result was computed
Taeuk-Jang/pointcompletion/pointnet/model.py
code served (permissive licence) · get_code("9de2aaf78bbf10fd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, STN3d: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np

torch.manual_seed(0)

# The STN3d.__init__ takes only 'device' as a parameter
# We'll use CPU device
device = 'cpu'

ARGS = {
    'device': device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type STN3d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a STN3d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STN3d Ran
ran on a synthesized input
hnmspirit/pointnet1/pointnet.py
pointer only (licence: NONE) · get_code("383dbc340bd8d282")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, hnmspirit/pointnet1 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

STN3d Ran
ran on a synthesized input
smfirmin/DeepLearningFinalProject/pointnet/model.py
code served (permissive licence) · get_code("daf339afe5dff1a4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

class STN3d(nn.Module):

    def __init__(self, device):
        super(STN3d, self).__init__()
        self.device = device
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, 9)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, x):
        """
        x: [batch_size, dimension, npoints]
        """
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)
        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)
        iden = torch.tensor([1, 0, 0, 0, 1, 0, 0, 0, 1], device=self.device).view(1, 9).repeat(batchsize, 1)
        x = x + iden
        x = x.view(-1, 3, 3)
        return x


INIT_ARGS = {
    'device': torch.device('cpu')
}

ARGS = {
    'x': torch.randn(2, 3, 5)
}

if __name__ == '__main__':
    model = STN3d(**INIT_ARGS)
    model.eval()
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 3, 3) — shape (2, 3, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
STN4d Ran
ran on a synthesized input · constructed an object; no result was computed
Taeuk-Jang/pointcompletion/pointnet/model.py
code served (permissive licence) · get_code("c51d2646e1f8707c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, STN4d: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.nn.functional as F

torch.manual_seed(0)

class STN4d(nn.Module):

    def __init__(self, device):
        super(STN4d, self).__init__()
        self.device = device
        self.conv1 = torch.nn.Conv1d(4, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, 12)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, x):
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)
        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)
        iden = Variable(torch.cat((torch.eye(3).repeat(batchsize, 1, 1), torch.zeros(batchsize, 1, 3)), dim=1))
        if x.is_cuda:
            iden = iden.to(device=self.device)
        x = x.view(-1, 4, 3)
        x = x + iden
        return x

device = torch.device('cpu')
stn4d_instance = STN4d(device=device)

ARGS = {
    'device': device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type STN4d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a STN4d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STNkd Ran
ran on a synthesized input · constructed an object; no result was computed
mengxingshifen1218/pointnet.pytorch/pointnet/model.py
code served (permissive licence) · get_code("b95fe13373b6da5e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, STNkd: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import logging
import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
import torch.nn.functional as F

torch.manual_seed(0)

class STNkd(nn.Module):

    def __init__(self, k=64):
        super(STNkd, self).__init__()
        self.conv1 = torch.nn.Conv1d(k, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k * k)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)
        self.k = k

    def forward(self, x):
        logging.info('The x0 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        logging.info('The x1 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = F.relu(self.bn2(self.conv2(x)))
        logging.info('The x2 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = F.relu(self.bn3(self.conv3(x)))
        logging.info('The x3 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = torch.max(x, 2, keepdim=True)[0]
        logging.info('The x4 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = x.view(-1, 1024)
        logging.info('The x5 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = F.relu(self.bn4(self.fc1(x)))
        logging.info('The x6 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = F.relu(self.bn5(self.fc2(x)))
        logging.info('The x7 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        x = self.fc3(x)
        logging.info('The x8 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        iden = Variable(torch.from_numpy(np.eye(self.k).flatten().astype(np.float32))).view(1, self.k * self.k).repeat(batchsize, 1)
        logging.info('The iden type is:%s,shape is:%s', str(type(iden)), str(iden.shape))
        logging.info(iden)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, self.k, self.k)
        logging.info('The x9 type is:%s,shape is:%s', str(type(x)), str(x.shape))
        return x


ARGS = {
    'k': 4
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type STNkd — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a STNkd output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STNkd Ran
ran on a synthesized input · constructed an object; no result was computed
Taeuk-Jang/pointcompletion/pointnet/model.py
code served (permissive licence) · get_code("bcbe4bef55f4d2d1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, STNkd: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
import torch.nn.functional as F

torch.manual_seed(0)

class STNkd(nn.Module):

    def __init__(self, k=64, device='cuda:1'):
        super(STNkd, self).__init__()
        self.device = device
        self.conv1 = torch.nn.Conv1d(k, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k * k)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)
        self.k = k

    def forward(self, x):
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)
        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)
        iden = Variable(torch.from_numpy(np.eye(self.k).flatten().astype(np.float32))).view(1, self.k * self.k).repeat(batchsize, 1)
        if x.is_cuda:
            iden = iden.to(device=self.device)
        x = x + iden
        x = x.view(-1, self.k, self.k)
        return x

# Create a small STNkd instance for testing
stn_module = STNkd(k=4, device='cpu')

# Create input tensor with shape (batch_size=2, k=4, num_points=3)
# The forward method expects input of shape (batch, k, num_points)
input_tensor = torch.randn(2, 4, 3)

ARGS = {
    'k': 4,
    'device': 'cpu'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type STNkd — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a STNkd output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STNkd Ran
ran on a synthesized input · constructed an object; no result was computed
princeton-vl/SimpleView/pointnet_pyt/pointnet/model.py
code served (permissive licence) · get_code("9198530155430cd2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, STNkd: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import numpy as np

torch.manual_seed(0)

# Create a tiny input tensor for the STNkd forward pass
# The model expects input of shape (batch_size, k, num_points)
# where k is the input feature dimension
batch_size = 2
k = 64
num_points = 10

# Create sample input tensor
x = torch.randn(batch_size, k, num_points)

ARGS = {
    'k': 64
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type STNkd — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a STNkd output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
STNkd Ran
ran on a synthesized input
smfirmin/DeepLearningFinalProject/pointnet/model.py
code served (permissive licence) · get_code("9002018eae6ef27f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

class STNkd(nn.Module):

    def __init__(self, device, k=64):
        super(STNkd, self).__init__()
        self.device = device
        self.conv1 = torch.nn.Conv1d(k, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k * k)
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)
        self.k = k

    def forward(self, x):
        """
        x: [batch_size, dimension, npoints]
        """
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)
        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)
        iden = torch.eye(self.k, device=self.device).flatten().view(1, self.k * self.k).repeat(batchsize, 1)
        x = x + iden
        x = x.view(-1, self.k, self.k)
        return x


INIT_ARGS = {
    'device': 'cpu',
    'k': 4
}

ARGS = {
    'x': torch.randn(2, 4, 3)
}

if __name__ == '__main__':
    model = STNkd(**INIT_ARGS)
    model.eval()
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 4, 4)")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 4, 4) — shape (2, 4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
TNet Ran
ran on a synthesized input · constructed an object; no result was computed
ftdlyc/pointnet_pytorch/models/pointnet.py
pointer only (licence: NONE) · get_code("58ce72343decd663")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, ftdlyc/pointnet_pytorch when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Tnet Ran
ran on a synthesized input · constructed an object; no result was computed
brendanmallery9/entropic-barycenters/classification_example/pointnet_implementation.py
code served (permissive licence) · get_code("9b8bdedaae9abf63")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Tnet: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Tnet only takes one parameter: k (number of input channels)
# Default is k=3, but we can test with a different value
k_value = 3

ARGS = {
    'k': k_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tnet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Tnet output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Tnet Ran
ran on a synthesized input
nikitakaraevv/pointnet/source/model.py
code served (permissive licence) · get_code("8de6ad0849c28980")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 3) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

class Tnet(nn.Module):

    def __init__(self, k=3):
        super().__init__()
        self.k = k
        self.conv1 = nn.Conv1d(k, 64, 1)
        self.conv2 = nn.Conv1d(64, 128, 1)
        self.conv3 = nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k * k)
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, input):
        bs = input.size(0)
        xb = F.relu(self.bn1(self.conv1(input)))
        xb = F.relu(self.bn2(self.conv2(xb)))
        xb = F.relu(self.bn3(self.conv3(xb)))
        pool = nn.MaxPool1d(xb.size(-1))(xb)
        flat = nn.Flatten(1)(pool)
        xb = F.relu(self.bn4(self.fc1(flat)))
        xb = F.relu(self.bn5(self.fc2(xb)))
        init = torch.eye(self.k, requires_grad=True).repeat(bs, 1, 1)
        if xb.is_cuda:
            init = init.cuda()
        matrix = self.fc3(xb).view(-1, self.k, self.k) + init
        return matrix


INIT_ARGS = {
    'k': 3
}

ARGS = {
    'input': torch.randn(2, 3, 8)
}

if __name__ == '__main__':
    model = Tnet(**INIT_ARGS)
    model.eval()
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 3, 3)")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 3, 3) — shape (2, 3, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
TransLayer Ran
ran on a synthesized input · constructed an object; no result was computed
yanxp/PointNet/model/pointnet.py
pointer only (licence: NONE) · get_code("d22507edb8c91af0")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, yanxp/PointNet when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Transform Ran
ran on a synthesized input · constructed an object; no result was computed
brendanmallery9/entropic-barycenters/classification_example/pointnet_implementation.py
code served (permissive licence) · get_code("cfb8085f652dcf26")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Transform: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# The Transform class has no __init__ parameters, so ARGS should be empty
ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Transform — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Transform output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Transform Ran
ran on a synthesized input
nikitakaraevv/pointnet/source/model.py
code served (permissive licence) · get_code("65aa5b5bf3ec8b52")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned tuple. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

class Tnet(nn.Module):

    def __init__(self, k=3):
        super().__init__()
        self.k = k
        self.conv1 = nn.Conv1d(k, 64, 1)
        self.conv2 = nn.Conv1d(64, 128, 1)
        self.conv3 = nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k * k)
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, input):
        bs = input.size(0)
        xb = F.relu(self.bn1(self.conv1(input)))
        xb = F.relu(self.bn2(self.conv2(xb)))
        xb = F.relu(self.bn3(self.conv3(xb)))
        pool = nn.MaxPool1d(xb.size(-1))(xb)
        flat = nn.Flatten(1)(pool)
        xb = F.relu(self.bn4(self.fc1(flat)))
        xb = F.relu(self.bn5(self.fc2(xb)))
        init = torch.eye(self.k, requires_grad=True).repeat(bs, 1, 1)
        if xb.is_cuda:
            init = init.cuda()
        matrix = self.fc3(xb).view(-1, self.k, self.k) + init
        return matrix

class Transform(nn.Module):

    def __init__(self):
        super().__init__()
        self.input_transform = Tnet(k=3)
        self.feature_transform = Tnet(k=64)
        self.conv1 = nn.Conv1d(3, 64, 1)
        self.conv2 = nn.Conv1d(64, 128, 1)
        self.conv3 = nn.Conv1d(128, 1024, 1)
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)

    def forward(self, input):
        matrix3x3 = self.input_transform(input)
        xb = torch.bmm(torch.transpose(input, 1, 2), matrix3x3).transpose(1, 2)
        xb = F.relu(self.bn1(self.conv1(xb)))
        matrix64x64 = self.feature_transform(xb)
        xb = torch.bmm(torch.transpose(xb, 1, 2), matrix64x64).transpose(1, 2)
        xb = F.relu(self.bn2(self.conv2(xb)))
        xb = self.bn3(self.conv3(xb))
        xb = nn.MaxPool1d(xb.size(-1))(xb)
        output = nn.Flatten(1)(xb)
        return (output, matrix3x3, matrix64x64)

INIT_ARGS = {}

ARGS = {
    'input': torch.randn(2, 3, 5)
}

if __name__ == '__main__':
    model = Transform(**INIT_ARGS)
    model.eval()
    output, matrix3x3, matrix64x64 = model(**ARGS)
    print("Output shape:", output.shape)
    print("Matrix 3x3 shape:", matrix3x3.shape)
    print("Matrix 64x64 shape:", matrix64x64.shape)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)

Showing the first 60. An agent gets every one from get_harvested_code_for_paper("1612.00593").

Also present in other papers' code

Of 164 samples lifted from this paper's linked repositories, 164 carry a function identity and 0 do not. Of the 153 distinct functions with an identity, 18 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 18 later) and 135 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
knn_point 21 other papers
21 later · 1 only as an assembly brick
2111.14819 (2021-11)
Point-BERT: Pre-training 3D Point Cloud Transformers with Masked Point Modeling
2202.07123 (2022-02)
Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework
2203.01137 (2022-03)
Self-Supervised Scene Flow Estimation with 4-D Automotive Radar
and 18 more not named here
conv3x3 18 other papers
18 later
1901.04780 (2019-01)
DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion
1905.09646 (2019-05)
Spatial Group-wise Enhance: Improving Semantic Feature Learning in Convolutional Networks
1910.10750 (2019-10)
6-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypoints
and 15 more not named here
reflection_matrix 14 other papers
14 later
1803.04996 (2018-03)
Comparing Task Simplifications to Learn Closed-Loop Object Picking Using Deep Reinforcement Learning
1805.01026 (2018-05)
Computing CNN Loss and Gradients for Pose Estimation with Riemannian Geometry
1807.00412 (2018-07)
Learning to Drive in a Day
and 11 more not named here
translation_from_matrix 14 other papers
14 later
1803.04996 (2018-03)
Comparing Task Simplifications to Learn Closed-Loop Object Picking Using Deep Reinforcement Learning
1805.01026 (2018-05)
Computing CNN Loss and Gradients for Pose Estimation with Riemannian Geometry
1807.00412 (2018-07)
Learning to Drive in a Day
and 11 more not named here
translation_matrix 14 other papers
14 later
1803.04996 (2018-03)
Comparing Task Simplifications to Learn Closed-Loop Object Picking Using Deep Reinforcement Learning
1805.01026 (2018-05)
Computing CNN Loss and Gradients for Pose Estimation with Riemannian Geometry
1807.00412 (2018-07)
Learning to Drive in a Day
and 11 more not named here
farthest_point_sample 9 other papers
9 later · 4 only as an assembly brick
1706.02413 (2017-06)
PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
2004.10904 (2020-04)
Through the Looking Glass: Neural 3D Reconstruction of Transparent Shapes
2103.10814 (2021-03)
Skeleton Merger: an Unsupervised Aligned Keypoint Detector
and 6 more not named here
conv2d
brick of this paper's assembly: helper, model
4 other papers
4 later
1803.05827 (2018-03)
Local Spectral Graph Convolution for Point Set Feature Learning
1809.07016 (2018-09)
Generating 3D Adversarial Point Clouds
2311.17243 (2023-11)
PHG-Net: Persistent Homology Guided Medical Image Classification
and 1 more not named here
feature_transform_regularizer 4 other papers
4 later
1905.06292 (2019-05)
3D Point Cloud Generative Adversarial Network Based on Tree Structured Graph Convolutions
1912.00280 (2019-12)
Morphing and Sampling Network for Dense Point Cloud Completion
2311.14265 (2023-11)
Adaptive Calibration: A Unified Conversion Framework of Spiking Neural Networks
and 1 more not named here
resnet18 4 other papers
4 later
1901.04780 (2019-01)
DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion
1910.10750 (2019-10)
6-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypoints
2003.00188 (2020-03)
Robust 6D Object Pose Estimation by Learning RGB-D Features
and 1 more not named here
resnet34 4 other papers
4 later
1901.04780 (2019-01)
DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion
1910.10750 (2019-10)
6-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypoints
2003.00188 (2020-03)
Robust 6D Object Pose Estimation by Learning RGB-D Features
and 1 more not named here
dropout
brick of this paper's assembly: helper, model
3 other papers
3 later · 1 only as an assembly brick
1801.07829 (2018-01)
Dynamic Graph CNN for Learning on Point Clouds
1807.09413 (2018-07)
3DFeat-Net: Weakly Supervised Local 3D Features for Point Cloud Registration
2208.01925 (2022-08)
SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud
max_pool2d
brick of this paper's assembly: helper, model
3 other papers
3 later · 1 only as an assembly brick
1801.07829 (2018-01)
Dynamic Graph CNN for Learning on Point Clouds
2204.11015 (2022-04)
Surface Reconstruction from Point Clouds by Learning Predictive Context Priors
2208.01925 (2022-08)
SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud
_variable_with_weight_decay
brick of this paper's assembly: helper
2 other papers
2 later · 1 only as an assembly brick
1801.07829 (2018-01)
Dynamic Graph CNN for Learning on Point Clouds
2208.01925 (2022-08)
SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud
batch_norm_template
brick of this paper's assembly: helper
2 other papers
2 later · 1 only as an assembly brick
1801.07829 (2018-01)
Dynamic Graph CNN for Learning on Point Clouds
2208.01925 (2022-08)
SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud
GridSamplingLayer 1 other paper
1 later
2107.09584 (2021-07)
Active 3D Shape Reconstruction from Vision and Touch
STN3d
brick of this paper's assembly: helper, model
1 other paper
1 later
2205.15225 (2022-05)
Few-shot Class-incremental Learning for 3D Point Cloud Objects
STNkd
brick of this paper's assembly: helper, model
1 other paper
1 later
2205.15225 (2022-05)
Few-shot Class-incremental Learning for 3D Point Cloud Objects
_variable_on_cpu
brick of this paper's assembly: helper
1 other paper
1 later
2305.15832 (2023-05)
All Points Matter: Entropy-Regularized Distribution Alignment for Weakly-supervised 3D Segmentation

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

More are linked than shown here.

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Point cloud is an important type of geometric data structure. Due to its irregular format, most researchers transform such data to regular 3D voxel grids or collections of images. This, however, renders data unnecessarily voluminous and causes issues. In this paper, we design a novel type of neural network that directly consumes point clouds, which well respects the permutation invariance of points in the input. Our network, named PointNet, provides a unified architecture for applications ranging from object classification, part segmentation, to scene semantic parsing. Though simple, PointNet is highly efficient and effective. Empirically, it shows strong performance on par or even better than state of the art. Theoretically, we provide analysis towards understanding of what the network has learnt and why the network is robust with respect to input perturbation and corruption.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("1612.00593")
get_code_for_paper("1612.00593")
have("1612.00593")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/1612.00593.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/1612.00593.svg)](https://syntology.ai/paper/1612.00593)

Connect an agent — have() is free.