SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1803.04249 · 2018

SO-Net: Self-Organizing Network for Point Cloud Analysis

arXiv · PDF · Open in the Atlas

Code that ran

7 of 7 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 5 have no contract verdict; 2 failures of our instrument; 0 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/1803.04249.json · badge: /api/ran/1803.04249.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
lijx10/SO-Net canonical (classifier label)
classifier basis: census
5 ran · 0 no recorded run
LONG-9621/SO-Net community (classifier label)
classifier basis not recorded on the link
2 ran · 0 no recorded run
FunctionStatusWhere it lives
compute_iou Ran
ran on a synthesized input
lijx10/SO-Net/models/losses.py
code served (permissive licence) · get_code("8f4466573c6f0ffe")
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 float64, ndim 0, 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 numpy as np
import json
import os
import tempfile
from collections import OrderedDict

torch.manual_seed(0)

# Create a temporary directory for the mock data
tmpdir = tempfile.mkdtemp()

# Create a mock part_color_mapping.json file
color_map = [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0], [1.0, 1.0, 0.0]]
color_map.extend([[0.5, 0.5, 0.5] for _ in range(46)])
color_map_file = os.path.join(tmpdir, 'part_color_mapping.json')
with open(color_map_file, 'w') as f:
    json.dump(color_map, f)

# Create mock score tensor: BxCxN where B=2, C=50 (num classes), N=8 (num points)
score = torch.randn(2, 50, 8)

# Create mock seg tensor: BxN where B=2, N=8 (ground truth segment labels)
seg = torch.randint(0, 50, (2, 8))

# Create mock label tensor: B where B=2 (object class labels, values 0-15)
label = torch.tensor([0, 5])

# Create mock input_pc tensor: Bx3xN where B=2, 3 coords, N=8 points
input_pc = torch.randn(2, 3, 8)

# Create a mock visualizer with display_current_results method
class MockVisualizer:
    def display_current_results(self, results_dict, epoch, total_iters):
        pass

visualizer = MockVisualizer()

# Create a mock opt with dataroot attribute
class MockOpt:
    def __init__(self, dataroot):
        self.dataroot = dataroot

opt = MockOpt(tmpdir)

ARGS = {
    'score': score,
    'seg': seg,
    'label': label,
    'visualizer': visualizer,
    'opt': opt,
    'input_pc': input_pc
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
compute_iou Ran
ran on a synthesized input
LONG-9621/SO-Net/models/losses.py
code served (permissive licence) · get_code("f1eb659f7c048f5c")
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 float64, ndim 0, 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 numpy as np
import json
import os
import tempfile
from collections import OrderedDict

torch.manual_seed(0)

# Create a temporary directory for the mock dataroot
temp_dir = tempfile.mkdtemp()

# Create a mock part_color_mapping.json file
color_map = [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0]] + [[0.5, 0.5, 0.5]] * 47
color_map_path = os.path.join(temp_dir, 'part_color_mapping.json')
with open(color_map_path, 'w') as f:
    json.dump(color_map, f)

# Create mock opt object with dataroot attribute
class MockOpt:
    def __init__(self, dataroot):
        self.dataroot = dataroot

opt = MockOpt(temp_dir)

# Create mock visualizer with display_current_results method
class MockVisualizer:
    def display_current_results(self, results_dict, epoch, total_epochs):
        pass

visualizer = MockVisualizer()

# Create tensors
# score: BxCxN tensor (batch=2, classes=50, points=4)
score = torch.randn(2, 50, 4)

# seg: BxN tensor (batch=2, points=4) - ground truth segmentation labels
seg = torch.randint(0, 50, (2, 4))

# label: batch labels indicating which object (0-15 for the 16 part categories)
label = torch.tensor([0, 5])

# input_pc: BxNx3 tensor (batch=2, points=4, coordinates=3)
input_pc = torch.randn(2, 4, 3)

ARGS = {
    'score': score,
    'seg': seg,
    'label': label,
    'visualizer': visualizer,
    'opt': opt,
    'input_pc': input_pc
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
compute_iou_np_array Ran
ran on a synthesized input
lijx10/SO-Net/models/losses.py
code served (permissive licence) · get_code("820fc38b598404a5")
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 1, dtype family float, shape (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
import numpy as np

torch.manual_seed(0)

# Analyze what the function uses:
# - score: torch tensor, used with torch.max(score, dim=1) and score.size()[0]
# - seg: torch tensor, used with indexing seg[i] and == comparisons
# - label: indexable (list or tensor), used with label[i] to index part_label
# - visualizer: not used in the function body
# - opt: not used in the function body
# - input_pc: not used in the function body

# Create small tensors for batch size 2
batch_size = 2
num_points = 4
num_classes = 40  # Based on part_label having 16 object categories with up to 40 total parts

# score: (batch_size, num_classes, num_points) or similar shape that supports dim=1 max
# Looking at usage: torch.max(score, dim=1) suggests classes are at dim=1
score = torch.randn(batch_size, num_classes, num_points)

# seg: (batch_size, num_points) with integer labels for parts (0-39)
seg = torch.randint(0, num_classes, (batch_size, num_points))

# label: (batch_size,) with integer indices into part_label (0-15 for 16 object categories)
label = torch.tensor([0, 5], dtype=torch.long)

# visualizer: not used, can be None or a dummy object
class DummyVisualizer:
    pass

visualizer = DummyVisualizer()

# opt: not used, can be None or a dummy object
class DummyOpt:
    pass

opt = DummyOpt()

# input_pc: not used, can be None or a dummy tensor
input_pc = torch.randn(batch_size, num_points, 3)

ARGS = {
    "score": score,
    "seg": seg,
    "label": label,
    "visualizer": visualizer,
    "opt": opt,
    "input_pc": input_pc
}
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 ndarray, ndim 1, dtype family float, shape (2,) — shape (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
compute_iou_np_array Ran
ran on a synthesized input
LONG-9621/SO-Net/models/losses.py
code served (permissive licence) · get_code("8dda5cb239b4da33")
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 1, dtype family float, shape (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
import numpy as np

torch.manual_seed(0)

# Analyze what the function actually uses:
# - score: torch tensor, used with torch.max(dim=1) and .size()[0]
# - seg: torch tensor, used with == operator and indexing [i]
# - label: indexable (list), used with label[i] to get indices into part_label
# - visualizer: not used in function body
# - opt: not used in function body
# - input_pc: not used in function body

# Build minimal valid arguments
batch_size = 2
num_points = 50
num_classes = 50

# score: (batch_size, num_classes, num_points) - logits for each class
score = torch.randn(batch_size, num_classes, num_points)

# seg: (batch_size, num_points) - ground truth segmentation labels (0-49)
seg = torch.randint(0, num_classes, (batch_size, num_points))

# label: (batch_size,) - object category labels (0-15, mapping to part_label indices)
label = torch.tensor([0, 5], dtype=torch.long).tolist()

# visualizer: not used, can be None
visualizer = None

# opt: not used, can be None or empty dict
opt = {}

# input_pc: not used, can be None
input_pc = None

ARGS = {
    "score": score,
    "seg": seg,
    "label": label,
    "visualizer": visualizer,
    "opt": opt,
    "input_pc": input_pc,
}
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 ndarray, ndim 1, dtype family float, shape (2,) — shape (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
knn_gather_by_indexing Ran
ran; our contract draft was wrong, not the code
lijx10/SO-Net/models/operations.py
code served (permissive licence) · get_code("501b125f7402f6d4")
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, 5) 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)

# Build tensors matching the function's expectations
B = 2  # batch size
C = 4  # channel/feature dimension
N = 3  # number of nodes
K = 5  # number of nearest neighbors

# som_node: BxCxN tensor
som_node = torch.randn(B, C, N)

# som_node_knn_I: BxNxK tensor of indices
# Indices should be in range [0, N) for gathering from dimension 2 of som_node
som_node_knn_I = torch.randint(0, N, (B, N, K))

ARGS = {
    "som_node": som_node,
    "som_node_knn_I": som_node_knn_I,
}
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, 5) — shape (2, 4, 3, 5) 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
knn_gather_wrapper Ran
ran on a synthesized input
lijx10/SO-Net/models/operations.py
code served (permissive licence) · get_code("46e06b4c07842f07")
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, 3, 4, 5) 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)

# Based on the function signature and body:
# knn_gather_wrapper expects:
# - som_node: BxCxN tensor where C is 2 or 3
# - som_node_knn_I: BxNxK tensor of indices

B = 2  # batch size
C = 3  # channel/feature dimension (must be 2 or 3 per assertion)
N = 4  # number of nodes
K = 5  # number of nearest neighbors

# Create som_node: BxCxN
som_node = torch.randn(B, C, N)

# Create som_node_knn_I: BxNxK with valid indices (0 to N-1)
som_node_knn_I = torch.randint(0, N, (B, N, K))

ARGS = {
    "som_node": som_node,
    "som_node_knn_I": som_node_knn_I,
}
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, 3, 4, 5) — shape (2, 3, 4, 5) 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
robust_norm Ran
ran; our contract draft was wrong, not the code
lijx10/SO-Net/models/losses.py
code served (permissive licence) · get_code("4a27074f82ccb58b")
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 float, shape (2, 4, 5) 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)

# The function expects a tensor of shape BxCxHxW
# It computes (var ** 2).sum(dim=2) which sums over the H dimension
# Result will be BxCxW
var = torch.randn(2, 4, 3, 5)  # B=2, C=4, H=3, W=5

ARGS = {
    "var": var
}
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 float, shape (2, 4, 5) — shape (2, 4, 5) 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

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 4 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 4 later) and 3 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
robust_norm 2 other papers
2 later
1904.12304 (2019-04)
RL-GAN-Net: A Reinforcement Learning Agent Controlled GAN Network for Real-Time Point Cloud Shape Completion
2007.00074 (2020-07)
Deep Geometric Texture Synthesis
compute_iou 1 other paper
1 later
1904.12304 (2019-04)
RL-GAN-Net: A Reinforcement Learning Agent Controlled GAN Network for Real-Time Point Cloud Shape Completion
compute_iou_np_array 1 other paper
1 later
1904.12304 (2019-04)
RL-GAN-Net: A Reinforcement Learning Agent Controlled GAN Network for Real-Time Point Cloud Shape Completion
knn_gather_by_indexing 1 other paper
1 later
2104.03501 (2021-04)
DeepI2P: Image-to-Point Cloud Registration via Deep Classification

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

This paper presents SO-Net, a permutation invariant architecture for deep learning with orderless point clouds. The SO-Net models the spatial distribution of point cloud by building a Self-Organizing Map (SOM). Based on the SOM, SO-Net performs hierarchical feature extraction on individual points and SOM nodes, and ultimately represents the input point cloud by a single feature vector. The receptive field of the network can be systematically adjusted by conducting point-to-node k nearest neighbor search. In recognition tasks such as point cloud reconstruction, classification, object part segmentation and shape retrieval, our proposed network demonstrates performance that is similar with or better than state-of-the-art approaches. In addition, the training speed is significantly faster than existing point cloud recognition networks because of the parallelizability and simplicity of the proposed architecture. Our code is available at the project website. https://github.com/lijx10/SO-Net

For agents

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

get_harvested_code_for_paper("1803.04249")
get_code_for_paper("1803.04249")
have("1803.04249")

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

curl https://syntology.ai/api/ran/1803.04249.json

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

[![Syntology run record](https://syntology.ai/api/ran/1803.04249.svg)](https://syntology.ai/paper/1803.04249)

Connect an agent — have() is free.