SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2103.07054 · CVPR · 2021

FS-Net: Fast Shape-based Network for Category-Level 6D Object Pose Estimation with Decoupled Rotation Mechanism

Wei Chen, Hyung Chang, Xi Jia, Linlin Shen, Jinming Duan, Aleš Leonardis

arXiv · PDF · Open in the Atlas

Code that ran

7 of 8 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (3 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 4 failures of our instrument; 1 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/2103.07054.json · badge: /api/ran/2103.07054.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
DC1991/FS_Net community
Papers with Code archive
7 ran · 1 no recorded run
FunctionStatusWhere it lives
Conv_layer Ran
ran on a synthesized input · constructed an object; no result was computed
DC1991/FS_Net/Net_archs.py
code served (permissive licence) · get_code("7e15bc064180c8f4")
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, Conv_layer: 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 torch.nn.functional as F
import math

torch.manual_seed(0)

# Define the helper functions and Conv_layer class as given
def indexing_neighbor(tensor, index):
    """
    Return: (bs, vertice_num, neighbor_num, dim)
    """
    bs, v, n = index.size()
    if bs == 1:
        tensor_indexed = tensor[torch.Tensor([[0]]).long(), index[0]].unsqueeze(dim=0)
    else:
        id_0 = torch.arange(bs).view(-1, 1, 1).long()
        tensor_indexed = tensor[id_0, index]
    return tensor_indexed

def get_neighbor_direction_norm(vertices, neighbor_index):
    """
    Return: (bs, vertice_num, neighobr_num, 3)
    """
    neighbors = indexing_neighbor(vertices, neighbor_index)
    neighbor_direction = neighbors - vertices.unsqueeze(2)
    neighbor_direction_norm = F.normalize(neighbor_direction, dim=-1)
    return neighbor_direction_norm.float()

class Conv_layer(nn.Module):

    def __init__(self, in_channel, out_channel, support_num):
        super().__init__()
        self.in_channel = in_channel
        self.out_channel = out_channel
        self.support_num = support_num
        self.relu = nn.ReLU(inplace=True)
        self.weights = nn.Parameter(torch.FloatTensor(in_channel, (support_num + 1) * out_channel))
        self.bias = nn.Parameter(torch.FloatTensor((support_num + 1) * out_channel))
        self.directions = nn.Parameter(torch.FloatTensor(3, support_num * out_channel))
        self.initialize()

    def initialize(self):
        stdv = 1.0 / math.sqrt(self.out_channel * (self.support_num + 1))
        self.weights.data.uniform_(-stdv, stdv)
        self.bias.data.uniform_(-stdv, stdv)
        self.directions.data.uniform_(-stdv, stdv)

    def forward(self, neighbor_index, vertices, feature_map):
        """
        Return: output feature map: (bs, vertice_num, out_channel)
        """
        bs, vertice_num, neighbor_num = neighbor_index.size()
        neighbor_direction_norm = get_neighbor_direction_norm(vertices, neighbor_index)
        support_direction_norm = F.normalize(self.directions, dim=0)
        theta = neighbor_direction_norm @ support_direction_norm
        theta = self.relu(theta)
        theta = theta.contiguous().view(bs, vertice_num, neighbor_num, -1)
        feature_out = feature_map @ self.weights + self.bias
        feature_center = feature_out[:, :, :self.out_channel]
        feature_support = feature_out[:, :, self.out_channel:]
        feature_support = indexing_neighbor(feature_support, neighbor_index)
        activation_support = theta * feature_support
        activation_support = activation_support.view(bs, vertice_num, neighbor_num, self.support_num, self.out_channel)
        activation_support = torch.max(activation_support, dim=2)[0]
        activation_support = torch.sum(activation_support, dim=2)
        feature_fuse = feature_center + activation_support
        return feature_fuse

# Build small test tensors and parameters
bs = 2  # batch size
vertice_num = 4  # number of vertices
neighbor_num = 3  # number of neighbors
in_channel = 5
out_channel = 6
support_num = 2

# Create neighbor_index: (bs, vertice_num, neighbor_num)
# Each vertex has neighbor_num neighbors sampled from vertice_num vertices
neighbor_index = torch.randint(0, vertice_num, (bs, vertice_num, neighbor_num))

# Create vertices: (bs, vertice_num, 3)
vertices = torch.randn(bs, vertice_num, 3)

# Create feature_map: (bs, vertice_num, in_channel)
feature_map = torch.randn(bs, vertice_num, in_channel)

# Create Conv_layer instance
conv_layer = Conv_layer(in_channel, out_channel, support_num)

ARGS = {
    'in_channel': in_channel,
    'out_channel': out_channel,
    'support_num': support_num,
}
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 Conv_layer — 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 Conv_layer 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
Conv_surface Ran
ran on a synthesized input · constructed an object; no result was computed
DC1991/FS_Net/Net_archs.py
code served (permissive licence) · get_code("7c76f8cfe4a9b93e")
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, Conv_surface: 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 torch.nn.functional as F
import math

torch.manual_seed(0)

def indexing_neighbor(tensor: '(bs, vertice_num, dim)', index: '(bs, vertice_num, neighbor_num)'):
    """
    Return: (bs, vertice_num, neighbor_num, dim)
    """
    bs, v, n = index.size()
    if bs == 1:
        tensor_indexed = tensor[torch.Tensor([[0]]).long(), index[0]].unsqueeze(dim=0)
    else:
        id_0 = torch.arange(bs).view(-1, 1, 1).long()
        tensor_indexed = tensor[id_0, index]
    return tensor_indexed

def get_neighbor_direction_norm(vertices: '(bs, vertice_num, 3)', neighbor_index: '(bs, vertice_num, neighbor_num)'):
    """
    Return: (bs, vertice_num, neighobr_num, 3)
    """
    neighbors = indexing_neighbor(vertices, neighbor_index)
    neighbor_direction = neighbors - vertices.unsqueeze(2)
    neighbor_direction_norm = F.normalize(neighbor_direction, dim=-1)
    return neighbor_direction_norm.float()

class Conv_surface(nn.Module):
    """Extract structure feafure from surface, independent from vertice coordinates"""

    def __init__(self, kernel_num, support_num):
        super().__init__()
        self.kernel_num = kernel_num
        self.support_num = support_num
        self.relu = nn.ReLU(inplace=True)
        self.directions = nn.Parameter(torch.FloatTensor(3, support_num * kernel_num))
        self.initialize()

    def initialize(self):
        stdv = 1.0 / math.sqrt(self.support_num * self.kernel_num)
        self.directions.data.uniform_(-stdv, stdv)

    def forward(self, neighbor_index: '(bs, vertice_num, neighbor_num)', vertices: '(bs, vertice_num, 3)'):
        """
        Return vertices with local feature: (bs, vertice_num, kernel_num)
        """
        bs, vertice_num, neighbor_num = neighbor_index.size()
        neighbor_direction_norm = get_neighbor_direction_norm(vertices, neighbor_index)
        support_direction_norm = F.normalize(self.directions, dim=0)
        theta = neighbor_direction_norm @ support_direction_norm
        theta = self.relu(theta)
        theta = theta.contiguous().view(bs, vertice_num, neighbor_num, self.support_num, self.kernel_num)
        theta = torch.max(theta, dim=2)[0]
        feature = torch.sum(theta, dim=2)
        return feature

# Create small tensors for testing
bs = 2
vertice_num = 4
neighbor_num = 3
kernel_num = 5
support_num = 2

# Create neighbor_index tensor
neighbor_index = torch.randint(0, vertice_num, (bs, vertice_num, neighbor_num))

# Create vertices tensor
vertices = torch.randn(bs, vertice_num, 3)

ARGS = {
    'kernel_num': kernel_num,
    'support_num': support_num
}
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 Conv_surface — 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 Conv_surface 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
Pool_layer Ran
ran on a synthesized input · constructed an object; no result was computed
DC1991/FS_Net/Net_archs.py
code served (permissive licence) · get_code("1da275fcc164e484")
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, Pool_layer: 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 = {
    'pooling_rate': 4,
    'neighbor_num': 4,
}
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 Pool_layer — 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 Pool_layer 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
get_nearest_index Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
DC1991/FS_Net/Net_archs.py
code served (permissive licence) · get_code("6c98740a684d1aad")
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 3, dtype family int, shape (2, 4, 1) 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 small tensors with distinct dimensions for clarity
# target: (bs=2, v1=4, 3)
# source: (bs=2, v2=5, 3)
target = torch.randn(2, 4, 3)
source = torch.randn(2, 5, 3)

ARGS = {
    'target': target,
    'source': source,
}
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 3, dtype family int, shape (2, 4, 1) — shape (2, 4, 1) 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
get_neighbor_direction_norm Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
DC1991/FS_Net/Net_archs.py
code served (permissive licence) · get_code("e9e5969423f92240")
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 4, dtype family float, shape (2, 4, 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.functional as F

torch.manual_seed(0)

# Create tiny tensors for testing
bs = 2  # batch size
vertice_num = 4  # number of vertices
neighbor_num = 3  # number of neighbors
dim = 3  # dimension for vertices (3D coordinates)

# vertices: (bs, vertice_num, 3)
vertices = torch.randn(bs, vertice_num, dim)

# neighbor_index: (bs, vertice_num, neighbor_num)
# Valid indices into vertice_num dimension (0-3)
neighbor_index = torch.tensor([
    [[0, 1, 2], [1, 2, 3], [0, 2, 3], [1, 2, 0]],
    [[1, 2, 3], [0, 1, 3], [0, 1, 2], [2, 3, 0]]
], dtype=torch.long)

ARGS = {
    'vertices': vertices,
    'neighbor_index': neighbor_index
}
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 4, dtype family float, shape (2, 4, 3, 3) — shape (2, 4, 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: none by the harness; any seed is in the setup block
get_neighbor_index Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
DC1991/FS_Net/Net_archs.py
code served (permissive licence) · get_code("e9fc6fa3d035121f")
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 3, dtype family int, shape (2, 4, 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 torch

torch.manual_seed(0)

# Create a small batch of vertices
# Shape: (bs=2, vertice_num=4, 3)
vertices = torch.randn(2, 4, 3)

neighbor_num = 2

ARGS = {
    "vertices": vertices,
    "neighbor_num": neighbor_num,
}
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 3, dtype family int, shape (2, 4, 2) — shape (2, 4, 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
indexing_neighbor Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
DC1991/FS_Net/Net_archs.py
code served (permissive licence) · get_code("3261de70d4811877")
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 4, dtype family float, shape (2, 4, 2, 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

torch.manual_seed(0)

# Create a tensor of shape (bs, vertice_num, dim)
# bs=2, vertice_num=4, dim=3
tensor = torch.randn(2, 4, 3)

# Create an index tensor of shape (bs, vertice_num, neighbor_num)
# bs=2, vertice_num=4, neighbor_num=2
# Each index should be in range [0, vertice_num)
index = torch.tensor([
    [[0, 1], [1, 2], [2, 3], [3, 0]],
    [[1, 2], [0, 3], [1, 0], [2, 1]]
], dtype=torch.long)

ARGS = {
    'tensor': tensor,
    'index': index
}
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 4, dtype family float, shape (2, 4, 2, 3) — shape (2, 4, 2, 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: none by the harness; any seed is in the setup block
GCN3D_segR Not yet run
unverified (harvested, no recorded run)
DC1991/FS_Net/Net_archs.py
code served (permissive licence) · get_code("df2669dbac03d994")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 2 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 2 later) and 6 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
get_nearest_index
brick of this paper's assembly: helper
1 other paper
1 later
2410.03644 (2024-10)
Unlearnable 3D Point Clouds: Class-wise Transformation Is All You Need
get_neighbor_index
brick of this paper's assembly: helper
1 other paper
1 later
2410.03644 (2024-10)
Unlearnable 3D Point Clouds: Class-wise Transformation Is All You Need

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

In this paper, we focus on category-level 6D pose and size estimation from a monocular RGB-D image. Previous methods suffer from inefficient category-level pose feature extraction, which leads to low accuracy and inference speed. To tackle this problem, we propose a fast shapebased network (FS-Net) with efficient category-level feature extraction for 6D pose estimation. First, we design an orientation aware autoencoder with 3D graph convolution for latent feature extraction. Thanks to the shift and scaleinvariance properties of 3D graph convolution, the learned latent feature is insensitive to point shift and object size. Then, to efficiently decode category-level rotation information from the latent feature, we propose a novel decoupled rotation mechanism that employs two decoders to complementarily access the rotation information. For translation and size, we estimate them by two residuals: the difference between the mean of object points and ground truth translation, and the difference between the mean size of the category and ground truth size, respectively. Finally, to increase the generalization ability of the FS-Net, we propose an online box-cage based 3D deformation mechanism to augment the training data. Extensive experiments on two benchmark datasets show that the proposed method achieves state-ofthe-art performance in both category-and instance-level 6D object pose estimation. Especially in category-level pose estimation, without extra synthetic data, our method outperforms existing methods by 6.3% on the NOCS-REAL dataset 1 .

For agents

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

get_harvested_code_for_paper("2103.07054")
get_code_for_paper("2103.07054")
have("2103.07054")

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

curl https://syntology.ai/api/ran/2103.07054.json

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

[![Syntology run record](https://syntology.ai/api/ran/2103.07054.svg)](https://syntology.ai/paper/2103.07054)

Connect an agent — have() is free.