SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2104.00678 · ICCV · 2021

Group-Free 3D Object Detection via Transformers

Zheng Zhang, Han Hu, Yue Cao, Xin Tong, Ze Liu

arXiv · PDF · Open in the Atlas

Code that ran

9 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (7 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 1 failure of our instrument; 8 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/2104.00678.json · badge: /api/ran/2104.00678.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
KookHoiKim/GroupFree3dBaseline community
Papers with Code archive
9 ran · 6 no recorded run
zeliu98/Group-Free-3D marked official in Papers with Code
Papers with Code archive
0 ran · 2 no recorded run
FunctionStatusWhere it lives
ClsAgnosticPredictHead Ran
ran on a synthesized input · constructed an object; no result was computed
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("1032f9a2b7eee988")
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, ClsAgnosticPredictHead: 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
import numpy as np

torch.manual_seed(0)

# Define the model class
class ClsAgnosticPredictHead(nn.Module):

    def __init__(self, num_class, num_heading_bin, num_proposal, seed_feat_dim=256):
        super().__init__()
        self.num_class = num_class
        self.num_heading_bin = num_heading_bin
        self.num_proposal = num_proposal
        self.seed_feat_dim = seed_feat_dim
        self.conv1 = torch.nn.Conv1d(seed_feat_dim, seed_feat_dim, 1)
        self.bn1 = torch.nn.BatchNorm1d(seed_feat_dim)
        self.conv2 = torch.nn.Conv1d(seed_feat_dim, seed_feat_dim, 1)
        self.bn2 = torch.nn.BatchNorm1d(seed_feat_dim)
        self.objectness_scores_head = torch.nn.Conv1d(seed_feat_dim, 1, 1)
        self.center_residual_head = torch.nn.Conv1d(seed_feat_dim, 3, 1)
        self.heading_class_head = torch.nn.Conv1d(seed_feat_dim, num_heading_bin, 1)
        self.heading_residual_head = torch.nn.Conv1d(seed_feat_dim, num_heading_bin, 1)
        self.size_pred_head = torch.nn.Conv1d(seed_feat_dim, 3, 1)
        self.sem_cls_scores_head = torch.nn.Conv1d(seed_feat_dim, self.num_class, 1)

    def forward(self, features, base_xyz, end_points, prefix=''):
        import torch.nn.functional as F
        batch_size = features.shape[0]
        num_proposal = features.shape[-1]
        net = F.relu(self.bn1(self.conv1(features)))
        net = F.relu(self.bn2(self.conv2(net)))
        objectness_scores = self.objectness_scores_head(net).transpose(2, 1)
        center_residual = self.center_residual_head(net).transpose(2, 1)
        center = base_xyz + center_residual
        heading_scores = self.heading_class_head(net).transpose(2, 1)
        heading_residuals_normalized = self.heading_residual_head(net).transpose(2, 1)
        heading_residuals = heading_residuals_normalized * (np.pi / self.num_heading_bin)
        pred_size = self.size_pred_head(net).transpose(2, 1).view([batch_size, num_proposal, 3])
        sem_cls_scores = self.sem_cls_scores_head(net).transpose(2, 1)
        end_points[f'{prefix}base_xyz'] = base_xyz
        end_points[f'{prefix}objectness_scores'] = objectness_scores
        end_points[f'{prefix}center'] = center
        end_points[f'{prefix}heading_scores'] = heading_scores
        end_points[f'{prefix}heading_residuals_normalized'] = heading_residuals_normalized
        end_points[f'{prefix}heading_residuals'] = heading_residuals
        end_points[f'{prefix}pred_size'] = pred_size
        end_points[f'{prefix}sem_cls_scores'] = sem_cls_scores
        return (center, pred_size)

# Create valid arguments
batch_size = 2
num_proposal = 5
seed_feat_dim = 4
num_class = 3
num_heading_bin = 6

features = torch.randn(batch_size, seed_feat_dim, num_proposal)
base_xyz = torch.randn(batch_size, num_proposal, 3)
end_points = {}
prefix = ''

ARGS = {
    'num_class': num_class,
    'num_heading_bin': num_heading_bin,
    'num_proposal': num_proposal,
    'seed_feat_dim': seed_feat_dim,
}
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 ClsAgnosticPredictHead — 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 ClsAgnosticPredictHead 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
FPSModule Ran
ran on a synthesized input · constructed an object; no result was computed
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("5a05b2fc118b751f")
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, FPSModule: 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 unittest.mock import MagicMock

torch.manual_seed(0)

# Mock the _ext module since it's a C++ extension
_ext = MagicMock()

# Create mock functions that return appropriate tensor shapes
def mock_furthest_point_sampling(xyz, npoint):
    B = xyz.shape[0]
    return torch.randint(0, xyz.shape[1], (B, npoint))

def mock_gather_points(features, idx):
    B, C, N = features.shape
    npoint = idx.shape[1]
    return torch.randn(B, C, npoint)

def mock_gather_points_grad(grad_out, idx, N):
    B, C = grad_out.shape[0], grad_out.shape[1]
    return torch.randn(B, C, N)

_ext.furthest_point_sampling = mock_furthest_point_sampling
_ext.gather_points = mock_gather_points
_ext.gather_points_grad = mock_gather_points_grad

# Define the functions and classes from the code
class FurthestPointSampling(torch.autograd.Function):
    @staticmethod
    def forward(ctx, xyz, npoint):
        fps_inds = _ext.furthest_point_sampling(xyz, npoint)
        ctx.mark_non_differentiable(fps_inds)
        return fps_inds

    @staticmethod
    def backward(xyz, a=None):
        return (None, None)

class GatherOperation(torch.autograd.Function):
    @staticmethod
    def forward(ctx, features, idx):
        _, C, N = features.size()
        ctx.for_backwards = (idx, C, N)
        return _ext.gather_points(features, idx)

    @staticmethod
    def backward(ctx, grad_out):
        idx, C, N = ctx.for_backwards
        grad_features = _ext.gather_points_grad(grad_out.contiguous(), idx, N)
        return (grad_features, None)

furthest_point_sample = FurthestPointSampling.apply
gather_operation = GatherOperation.apply

class FPSModule(nn.Module):
    def __init__(self, num_proposal):
        super().__init__()
        self.num_proposal = num_proposal

    def forward(self, xyz, features):
        sample_inds = furthest_point_sample(xyz, self.num_proposal)
        xyz_flipped = xyz.transpose(1, 2).contiguous()
        new_xyz = gather_operation(xyz_flipped, sample_inds).transpose(1, 2).contiguous()
        new_features = gather_operation(features, sample_inds).contiguous()
        return (new_xyz, new_features, sample_inds)

# Build setup arguments
ARGS = {
    'num_proposal': 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 FPSModule — 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 FPSModule 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
PointnetFPModule Ran
ran on a synthesized input · constructed an object; no result was computed
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("d53ad68b74734b36")
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, PointnetFPModule: 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 typing import List

torch.manual_seed(0)

ARGS = {
    'mlp': [64, 128, 256],
    'bn': 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 PointnetFPModule — 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 PointnetFPModule 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
PointsObjClsModule Ran
ran on a synthesized input · constructed an object; no result was computed
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("8a069336f892a257")
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, PointsObjClsModule: 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 function only uses seed_feature_dim parameter
# seed_feature_dim should be an integer representing the feature dimension
seed_feature_dim = 4

ARGS = {
    'seed_feature_dim': seed_feature_dim
}

# Verify by instantiating and running forward pass
module = nn.Module.__new__(type('PointsObjClsModule', (nn.Module,), {
    '__init__': lambda self, seed_feature_dim: (
        super(type(self), self).__init__(),
        setattr(self, 'in_dim', seed_feature_dim),
        setattr(self, 'conv1', torch.nn.Conv1d(self.in_dim, self.in_dim, 1)),
        setattr(self, 'bn1', torch.nn.BatchNorm1d(self.in_dim)),
        setattr(self, 'conv2', torch.nn.Conv1d(self.in_dim, self.in_dim, 1)),
        setattr(self, 'bn2', torch.nn.BatchNorm1d(self.in_dim)),
        setattr(self, 'conv3', torch.nn.Conv1d(self.in_dim, 1, 1))
    )[-1],
    'forward': lambda self, seed_features: (
        torch.nn.functional.relu(self.bn1(self.conv1(seed_features))),
        torch.nn.functional.relu(self.bn2(self.conv2(torch.nn.functional.relu(self.bn1(self.conv1(seed_features)))))),
        self.conv3(torch.nn.functional.relu(self.bn2(self.conv2(torch.nn.functional.relu(self.bn1(self.conv1(seed_features)))))))
    )[-1]
}))

# Create a proper instance using the actual class definition
class PointsObjClsModule(nn.Module):
    def __init__(self, seed_feature_dim):
        super().__init__()
        self.in_dim = seed_feature_dim
        self.conv1 = torch.nn.Conv1d(self.in_dim, self.in_dim, 1)
        self.bn1 = torch.nn.BatchNorm1d(self.in_dim)
        self.conv2 = torch.nn.Conv1d(self.in_dim, self.in_dim, 1)
        self.bn2 = torch.nn.BatchNorm1d(self.in_dim)
        self.conv3 = torch.nn.Conv1d(self.in_dim, 1, 1)

    def forward(self, seed_features):
        net = torch.nn.functional.relu(self.bn1(self.conv1(seed_features)))
        net = torch.nn.functional.relu(self.bn2(self.conv2(net)))
        logits = self.conv3(net)
        return logits

# Test instantiation
model = PointsObjClsModule(**ARGS)

# Test forward pass with appropriate tensor shape
batch_size = 2
num_seed = 5
seed_features = torch.randn(batch_size, seed_feature_dim, num_seed)
output = model(seed_features)

assert output.shape == (batch_size, 1, num_seed), f"Expected shape {(batch_size, 1, num_seed)}, got {output.shape}"
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 PointsObjClsModule — 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 PointsObjClsModule 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
PositionEmbeddingLearned Ran
ran on a synthesized input · constructed an object; no result was computed
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("490c93aa45d7780a")
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, PositionEmbeddingLearned: 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)

# Analyze the PositionEmbeddingLearned class:
# - __init__ takes: input_channel, num_pos_feats=288
# - forward takes: xyz
# - xyz is transposed from (batch, num_pos_feats, num_points) to (batch, input_channel, num_points)
# - So input xyz shape should be (batch, num_points, input_channel)

# Create small test input
batch_size = 2
num_points = 4
input_channel = 3

# xyz tensor: (batch, num_points, input_channel)
xyz = torch.randn(batch_size, num_points, input_channel)

ARGS = {
    "input_channel": input_channel,
    "num_pos_feats": 288,
}
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 PositionEmbeddingLearned — 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 PositionEmbeddingLearned 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
PredictHead Ran
ran on a synthesized input
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("e0691173a4ce4a26")
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 for PredictHead.__init__
num_class = 5
num_heading_bin = 6
num_size_cluster = 4
num_proposal = 8
seed_feat_dim = 16

# Create mean_size_arr: shape (num_size_cluster, 3)
mean_size_arr = np.array([
    [1.0, 2.0, 1.5],
    [2.0, 3.0, 2.5],
    [1.5, 2.5, 2.0],
    [3.0, 4.0, 3.5]
], dtype=np.float32)

INIT_ARGS = {
    'num_class': num_class,
    'num_heading_bin': num_heading_bin,
    'num_size_cluster': num_size_cluster,
    'mean_size_arr': mean_size_arr,
    'num_proposal': num_proposal,
    'seed_feat_dim': seed_feat_dim
}

# ARGS for PredictHead.forward
batch_size = 2

# features: (B, C, num_proposal)
features = torch.randn(batch_size, seed_feat_dim, num_proposal)

# base_xyz: (B, num_proposal, 3)
base_xyz = torch.randn(batch_size, num_proposal, 3)

# end_points: dictionary to store outputs
end_points = {}

# prefix: string prefix for keys in end_points
prefix = 'test_'

ARGS = {
    'features': features,
    'base_xyz': base_xyz,
    'end_points': end_points,
    'prefix': prefix
}
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)
ThreeInterpolate Ran
ran on a synthesized input · constructed an object; no result was computed
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("565863bfed9f02fc")
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, ThreeInterpolate: 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 mock extension module with the necessary functions
class MockExt:
    @staticmethod
    def three_interpolate(features, idx, weight):
        """Mock implementation that returns interpolated features"""
        B, c, n = features.size(0), features.size(1), idx.size(1)
        return torch.zeros(B, c, n)
    
    @staticmethod
    def three_interpolate_grad(grad_out, idx, weight, m):
        """Mock implementation that returns gradients for features"""
        B, c = grad_out.size(0), grad_out.size(1)
        return torch.zeros(B, c, m)

# Replace the _ext module reference with our mock
_ext = MockExt()

# Create valid arguments for ThreeInterpolate
B = 2  # batch size
c = 4  # number of channels/features
m = 5  # number of source points
n = 3  # number of target points

# features: (B, c, m) - features to interpolate from
features = torch.randn(B, c, m)

# idx: (B, n, 3) - indices of 3 nearest neighbors for each target point
# Values should be valid indices in range [0, m)
idx = torch.randint(0, m, (B, n, 3))

# weight: (B, n, 3) - interpolation weights (should sum to 1 for each target point)
weight = torch.ones(B, n, 3) / 3.0

ARGS = {
    "ctx": None,  # ctx will be provided by autograd
    "features": features,
    "idx": idx,
    "weight": weight
}
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 ThreeInterpolate — 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 ThreeInterpolate 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
ThreeNN Ran
ran on a synthesized input · constructed an object; no result was computed
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("11f8d19732aa2254")
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, ThreeNN: 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 sys

# Create a mock _ext module with three_nn function since it's a C++ extension
class MockExt:
    @staticmethod
    def three_nn(unknown, known):
        """Mock implementation of three_nn"""
        B, n, _ = unknown.shape
        B, m, _ = known.shape
        
        # Compute pairwise squared distances
        # unknown: (B, n, 3), known: (B, m, 3)
        dist2 = torch.cdist(unknown, known) ** 2  # (B, n, m)
        
        # Get 3 nearest neighbors (or fewer if m < 3)
        k = min(3, m)
        dist2_sorted, idx = torch.topk(dist2, k, dim=2, largest=False)
        
        # Pad to (B, n, 3) if needed
        if k < 3:
            pad_size = 3 - k
            dist2_sorted = torch.nn.functional.pad(dist2_sorted, (0, pad_size), value=float('inf'))
            idx = torch.nn.functional.pad(idx, (0, pad_size), value=0)
        
        return dist2_sorted, idx

# Mock the _ext module
sys.modules['_ext'] = MockExt()
_ext = MockExt()

torch.manual_seed(0)

# Create small test tensors
# unknown: (B, n, 3) - B=2 batch, n=4 unknown points, 3 features
# known: (B, m, 3) - B=2 batch, m=5 known points, 3 features
unknown = torch.randn(2, 4, 3)
known = torch.randn(2, 5, 3)

ARGS = {
    'unknown': unknown,
    'known': known
}
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 ThreeNN — 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 ThreeNN 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
multi_head_attention_forward Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("89a5e4ca75931eac")
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 tuple. 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
import torch.nn.functional as F

torch.manual_seed(0)

# Set up dimensions
embed_dim = 8
num_heads = 2
head_dim = embed_dim // num_heads
batch_size = 2
tgt_len = 3
src_len = 4

# Create query, key, value tensors
query = torch.randn(tgt_len, batch_size, embed_dim)
key = torch.randn(src_len, batch_size, embed_dim)
value = torch.randn(src_len, batch_size, embed_dim)

# Create projection weights and biases
# in_proj_weight combines q, k, v projection weights: shape (3*embed_dim, embed_dim)
in_proj_weight = torch.randn(3 * embed_dim, embed_dim)
in_proj_bias = torch.randn(3 * embed_dim)

# Output projection weights and biases
out_proj_weight = torch.randn(embed_dim, embed_dim)
out_proj_bias = torch.randn(embed_dim)

# Optional bias_k and bias_v (can be None or tensors)
bias_k = None
bias_v = None

# Optional masks
key_padding_mask = None
attn_mask = None

# Optional separate projection weights (used when use_separate_proj_weight=True)
q_proj_weight = None
k_proj_weight = None
v_proj_weight = None

# Optional static k and v
static_k = None
static_v = None

ARGS = {
    'query': query,
    'key': key,
    'value': value,
    'embed_dim_to_check': embed_dim,
    'num_heads': num_heads,
    'in_proj_weight': in_proj_weight,
    'in_proj_bias': in_proj_bias,
    'bias_k': bias_k,
    'bias_v': bias_v,
    'add_zero_attn': False,
    'dropout_p': 0.0,
    'out_proj_weight': out_proj_weight,
    'out_proj_bias': out_proj_bias,
    'training': True,
    'key_padding_mask': key_padding_mask,
    'need_weights': True,
    'attn_mask': attn_mask,
    'use_separate_proj_weight': False,
    'q_proj_weight': q_proj_weight,
    'k_proj_weight': k_proj_weight,
    'v_proj_weight': v_proj_weight,
    'static_k': static_k,
    'static_v': static_v,
}
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 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: none by the harness; any seed is in the setup block
GeneralSamplingModule Not yet run
unverified (harvested, no recorded run)
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("ba5e770e929c3505")
How it ran

no invocation recorded.

GroupFreeDetector Not yet run
unverified (harvested, no recorded run)
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("8a951c82c3d19407")
How it ran

no invocation recorded.

GroupFreeDetector Not yet run
unverified (harvested, no recorded run)
zeliu98/Group-Free-3D/models/detector.py
code served (permissive licence) · get_code("ddd521a89f668ed8")
How it ran

no invocation recorded.

MultiheadAttention Not yet run
unverified (harvested, no recorded run)
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("6bd1a32eae6a6a53")
How it ran

no invocation recorded.

Pointnet2Backbone Not yet run
unverified (harvested, no recorded run)
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("dba99ebc3fa7f7a2")
How it ran

no invocation recorded.

Pointnet2Backbone Not yet run
unverified (harvested, no recorded run)
zeliu98/Group-Free-3D/models/detector.py
code served (permissive licence) · get_code("b3d2771d692667b7")
How it ran

no invocation recorded.

PointnetSAModuleVotes Not yet run
unverified (harvested, no recorded run)
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("ee17b6fa95e4d8fa")
How it ran

no invocation recorded.

TransformerDecoderLayer Not yet run
unverified (harvested, no recorded run)
KookHoiKim/GroupFree3dBaseline/models/detector.py
code served (permissive licence) · get_code("582548d08ea83826")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 6 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 6 later) and 11 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
PointnetSAModuleVotes
brick of this paper's assembly: helper
4 other papers
1 dated earlier, 3 later · 3 only as an assembly brick
1706.02413 (2017-06)
PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
2204.13630 (2022-04)
Rotationally Equivariant 3D Object Detection
2301.02508 (2023-01)
End-to-End 3D Dense Captioning with Vote2Cap-DETR
and 1 more not named here
ThreeInterpolate
brick of this paper's assembly: helper
4 other papers
1 dated earlier, 3 later · 4 only as an assembly brick
1706.02413 (2017-06)
PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
2204.13630 (2022-04)
Rotationally Equivariant 3D Object Detection
2207.05444 (2022-07)
Category-Level 6D Object Pose and Size Estimation using Self-Supervised Deep Prior Deformation Networks
and 1 more not named here
ThreeNN
brick of this paper's assembly: helper
4 other papers
1 dated earlier, 3 later · 4 only as an assembly brick
1706.02413 (2017-06)
PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
2204.13630 (2022-04)
Rotationally Equivariant 3D Object Detection
2207.05444 (2022-07)
Category-Level 6D Object Pose and Size Estimation using Self-Supervised Deep Prior Deformation Networks
and 1 more not named here
PointnetFPModule
brick of this paper's assembly: helper
3 other papers
2 dated earlier, 1 later · 2 only as an assembly brick
1706.02413 (2017-06)
PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
2009.12606 (2020-09)
Grasp Proposal Networks: An End-to-End Solution for Visual Learning of Robotic Grasps
2204.13630 (2022-04)
Rotationally Equivariant 3D Object Detection
multi_head_attention_forward
brick of this paper's assembly: helper
3 other papers
3 later
2203.11496 (2022-03)
TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection with Transformers
2403.15241 (2024-03)
IS-FUSION: Instance-Scene Collaborative Fusion for Multimodal 3D Object Detection
2503.14493 (2025-03)
State Space Model Meets Transformer: A New Paradigm for 3D Object Detection
MultiheadAttention
brick of this paper's assembly: model
1 other paper
1 later
2105.13868 (2021-05)
Learning Relation Alignment for Calibrated Cross-modal Retrieval

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

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

Abstract

Recently, directly detecting 3D objects from 3D point clouds has received increasing attention. To extract object representation from an irregular point cloud, existing methods usually take a point grouping step to assign the points to an object candidate so that a PointNet-like network could be used to derive object features from the grouped points. However, the inaccurate point assignments caused by the hand-crafted grouping scheme decrease the performance of 3D object detection. In this paper, we present a simple yet effective method for directly detecting 3D objects from the 3D point cloud. Instead of grouping local points to each object candidate, our method computes the feature of an object from all the points in the point cloud with the help of an attention mechanism in the Transformers [42], where the contribution of each point is automatically learned in the network training. With an improved attention stacking scheme, our method fuses object features in different stages and generates more accurate object detection results. With few bells and whistles, the proposed method achieves state-of-the-art 3D object detection performance on two widely used benchmarks, Scan-Net V2 and SUN RGB-D. The code and models are publicly available at https://github.com/zeliu98/ Group-Free-3D

For agents

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

get_harvested_code_for_paper("2104.00678")
get_code_for_paper("2104.00678")
have("2104.00678")

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

curl https://syntology.ai/api/ran/2104.00678.json

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

[![Syntology run record](https://syntology.ai/api/ran/2104.00678.svg)](https://syntology.ai/paper/2104.00678)

Connect an agent — have() is free.