SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1904.08889 · 2019

KPConv: Flexible and Deformable Convolution for Point Clouds

arXiv · PDF · Open in the Atlas

Code that ran

9 of 12 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; 1 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 5 failures of our instrument; 3 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/1904.08889.json · badge: /api/ran/1904.08889.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
HuguesTHOMAS/KPConv canonical (classifier label)
classifier basis not recorded on the link
4 ran · 2 no recorded run
isl-org/Open3D-ML extension (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
HuguesTHOMAS/KPConv-PyTorch reimplementation (classifier label)
classifier basis: classifier_2025
2 ran · 1 no recorded run
FunctionStatusWhere it lives
IoU_class_metrics Ran
ran on a synthesized input
HuguesTHOMAS/KPConv/plot_convergence.py
code served (permissive licence) · get_code("814a2fe9241bee1e")
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 numpy as np

# Construct valid arguments for IoU_class_metrics
# The function expects:
# - all_IoUs: a list of arrays where each array contains IoU scores per class
# - smooth_n: an integer for smoothing window size

# Create a list of IoU arrays (one per epoch)
# Each array represents IoU scores for multiple classes
all_IoUs = [
    np.array([[0.7, 0.8, 0.75],      # epoch 0: 3 classes
              [0.65, 0.82, 0.78]]),   
    np.array([[0.72, 0.81, 0.76],    # epoch 1: 3 classes
              [0.68, 0.83, 0.79]]),
    np.array([[0.74, 0.82, 0.77],    # epoch 2: 3 classes
              [0.70, 0.84, 0.80]]),
    np.array([[0.76, 0.83, 0.78],    # epoch 3: 3 classes
              [0.72, 0.85, 0.81]]),
]

smooth_n = 1

ARGS = {
    "all_IoUs": all_IoUs,
    "smooth_n": smooth_n,
}
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 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
IoU_multi_metrics Ran
ran on a synthesized input
HuguesTHOMAS/KPConv/plot_convergence.py
code served (permissive licence) · get_code("09d5696850a35aee")
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 numpy as np

# Analyze the function:
# - all_IoUs: list of epochs, each epoch contains list of objects, each object has IoUs (2D array)
# - smooth_n: integer for smoothing window
# The function expects nested structure: all_IoUs[epoch][object] = 2D array of IoUs

# Create sample IoU data
# Structure: all_IoUs[epoch_idx][object_idx] = 2D array of shape (num_classes, num_samples)

# 3 epochs
epoch_1_obj_1 = np.random.rand(3, 4)  # 3 classes, 4 samples
epoch_1_obj_2 = np.random.rand(3, 5)  # 3 classes, 5 samples
epoch_1 = [epoch_1_obj_1, epoch_1_obj_2]

epoch_2_obj_1 = np.random.rand(3, 4)
epoch_2_obj_2 = np.random.rand(3, 5)
epoch_2 = [epoch_2_obj_1, epoch_2_obj_2]

epoch_3_obj_1 = np.random.rand(3, 4)
epoch_3_obj_2 = np.random.rand(3, 5)
epoch_3 = [epoch_3_obj_1, epoch_3_obj_2]

all_IoUs = [epoch_1, epoch_2, epoch_3]

smooth_n = 1

ARGS = {
    "all_IoUs": all_IoUs,
    "smooth_n": smooth_n,
}
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 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
bias_variable Ran
ran; our contract draft was wrong, not the code
HuguesTHOMAS/KPConv/models/network_blocks.py
code served (permissive licence) · get_code("6a88f1ae24fa26b2")
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 ResourceVariable, ndim 1, dtype family float, shape (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 tensorflow as tf

def bias_variable(shape):
    initial = tf.constant(0.0, shape=shape)
    return tf.Variable(initial, name='bias')

# The function takes a single parameter 'shape' which should be a tuple or list
# defining the shape of the bias variable. We'll create a small shape tuple.

ARGS = {
    'shape': (4,)
}
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 ResourceVariable, ndim 1, dtype family float, shape (4,) — shape (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: none by the harness; any seed is in the setup block
closest_pool Ran
ran; our contract draft was wrong, not the code
isl-org/Open3D-ML/ml3d/torch/models/kpconv.py
pointer only (licence: NOASSERTION) · get_code("bbafdd82ae3987bf")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, isl-org/Open3D-ML (NOASSERTION). 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.

create_3D_rotations Ran
ran, honoured the contract we drafted
HuguesTHOMAS/KPConv-PyTorch/kernels/kernel_points.py
code served (permissive licence) · get_code("c545a5b0a698e295")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned ndarray, 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 numpy as np

# Create sample data for the function
# axis: float32[N, 3] - N rotation axes (should be unit vectors)
# angle: float32[N,] - N rotation angles in radians

N = 2

# Create normalized axis vectors
axis = np.array([
    [1.0, 0.0, 0.0],  # rotation around x-axis
    [0.0, 1.0, 0.0]   # rotation around y-axis
], dtype=np.float32)

# Create angles in radians
angle = np.array([np.pi / 4, np.pi / 6], dtype=np.float32)

ARGS = {
    'axis': axis,
    'angle': angle
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
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, 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: none by the harness; any seed is in the setup block
gather Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
isl-org/Open3D-ML/ml3d/torch/models/kpconv.py
pointer only (licence: NOASSERTION) · get_code("9a89c500824a1c5c")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, isl-org/Open3D-ML (NOASSERTION). 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.

kernel_point_optimization_debug Ran
ran; our contract draft was wrong, not the code
HuguesTHOMAS/KPConv-PyTorch/kernels/kernel_points.py
code served (permissive licence) · get_code("a40d478bb51450e9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) 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 numpy as np
import torch

torch.manual_seed(0)
np.random.seed(0)

# The function performs kernel point optimization for point cloud convolutions
# Parameters:
# - radius: scalar, radius of the kernels
# - num_points: int, number of points per kernel
# - num_kernels: int, number of kernels to create
# - dimension: int, spatial dimension (2D or 3D typically)
# - fixed: str, which points to fix ('none', 'center', 'verticals')
# - ratio: float, ratio of radius for final placement
# - verbose: int, verbosity level

ARGS = {
    'radius': 0.5,
    'num_points': 5,
    'num_kernels': 2,
    'dimension': 3,
    'fixed': 'center',
    'ratio': 0.66,
    'verbose': 0
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
matplotlib, 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: none by the harness; any seed is in the setup block
radius_gaussian Ran
ran; our contract draft was wrong, not the code
isl-org/Open3D-ML/ml3d/torch/models/kpconv.py
pointer only (licence: NOASSERTION) · get_code("c0186f8ddb0bc115")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, isl-org/Open3D-ML (NOASSERTION). 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.

running_mean Ran
ran on a synthesized input
HuguesTHOMAS/KPConv/plot_convergence.py
code served (permissive licence) · get_code("2e38027f9a1a89e6")
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 (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 numpy as np

# Test case 1: 1D signal
signal_1d = np.array([1.0, 2.0, 3.0, 4.0, 5.0])
n_value = 1

# Test case 2: 2D signal with axis=0
signal_2d = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [5.0, 6.0, 7.0, 8.0]
])

ARGS = {
    "signal": signal_1d,
    "n": n_value,
    "axis": 0
}
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 1, dtype family float, shape (5,) — shape (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
ind_max_pool Not yet run
unverified (harvested, no recorded run)
HuguesTHOMAS/KPConv/models/network_blocks.py
code served (permissive licence) · get_code("4383ecd8cc0719fa")
How it ran

no invocation recorded.

p2p_fitting_regularizer Not yet run
unverified (harvested, no recorded run)
HuguesTHOMAS/KPConv-PyTorch/models/architectures.py
code served (permissive licence) · get_code("9b6e188d38ea6880")
How it ran

no invocation recorded.

weight_variable Not yet run
unverified (harvested, no recorded run)
HuguesTHOMAS/KPConv/models/network_blocks.py
code served (permissive licence) · get_code("5a6188c7d118c3fc")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 10 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 10 later) and 2 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
gather 8 other papers
8 later
2007.12668 (2020-07)
KPRNet: Improving projection-based LiDAR semantic segmentation
2111.12591 (2021-11)
Lepard: Learning partial point cloud matching in rigid and deformable scenes
2203.04041 (2022-03)
Shape-invariant 3D Adversarial Point Clouds
and 5 more not named here
radius_gaussian 8 other papers
8 later
2007.12668 (2020-07)
KPRNet: Improving projection-based LiDAR semantic segmentation
2111.12591 (2021-11)
Lepard: Learning partial point cloud matching in rigid and deformable scenes
2203.04041 (2022-03)
Shape-invariant 3D Adversarial Point Clouds
and 5 more not named here
closest_pool 6 other papers
6 later
2007.12668 (2020-07)
KPRNet: Improving projection-based LiDAR semantic segmentation
2111.12591 (2021-11)
Lepard: Learning partial point cloud matching in rigid and deformable scenes
2203.04041 (2022-03)
Shape-invariant 3D Adversarial Point Clouds
and 3 more not named here
bias_variable 4 other papers
4 later
2003.03164 (2020-03)
D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features
2003.13035 (2020-03)
Multi-Path Region Mining For Weakly Supervised 3D Semantic Segmentation on Point Clouds
2007.06888 (2020-07)
JSENet: Joint Semantic Segmentation and Edge Detection Network for 3D Point Clouds
and 1 more not named here
create_3D_rotations 4 other papers
4 later
2007.12668 (2020-07)
KPRNet: Improving projection-based LiDAR semantic segmentation
2301.10222 (2023-01)
RangeViT: Towards Vision Transformers for 3D Semantic Segmentation in Autonomous Driving
and 1 more not named here
ind_max_pool 4 other papers
4 later
2003.03164 (2020-03)
D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features
2003.13035 (2020-03)
Multi-Path Region Mining For Weakly Supervised 3D Semantic Segmentation on Point Clouds
2007.06888 (2020-07)
JSENet: Joint Semantic Segmentation and Edge Detection Network for 3D Point Clouds
and 1 more not named here
weight_variable 4 other papers
4 later
2003.03164 (2020-03)
D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features
2003.13035 (2020-03)
Multi-Path Region Mining For Weakly Supervised 3D Semantic Segmentation on Point Clouds
2007.06888 (2020-07)
JSENet: Joint Semantic Segmentation and Edge Detection Network for 3D Point Clouds
and 1 more not named here
IoU_class_metrics 1 other paper
1 later
2003.13035 (2020-03)
Multi-Path Region Mining For Weakly Supervised 3D Semantic Segmentation on Point Clouds
IoU_multi_metrics 1 other paper
1 later
2003.13035 (2020-03)
Multi-Path Region Mining For Weakly Supervised 3D Semantic Segmentation on Point Clouds
running_mean 1 other paper
1 later
2003.13035 (2020-03)
Multi-Path Region Mining For Weakly Supervised 3D Semantic Segmentation on Point Clouds

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

We present Kernel Point Convolution (KPConv), a new design of point convolution, i.e. that operates on point clouds without any intermediate representation. The convolution weights of KPConv are located in Euclidean space by kernel points, and applied to the input points close to them. Its capacity to use any number of kernel points gives KPConv more flexibility than fixed grid convolutions. Furthermore, these locations are continuous in space and can be learned by the network. Therefore, KPConv can be extended to deformable convolutions that learn to adapt kernel points to local geometry. Thanks to a regular subsampling strategy, KPConv is also efficient and robust to varying densities. Whether they use deformable KPConv for complex tasks, or rigid KPconv for simpler tasks, our networks outperform state-of-the-art classification and segmentation approaches on several datasets. We also offer ablation studies and visualizations to provide understanding of what has been learned by KPConv and to validate the descriptive power of deformable KPConv.

For agents

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

get_harvested_code_for_paper("1904.08889")
get_code_for_paper("1904.08889")
have("1904.08889")

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

curl https://syntology.ai/api/ran/1904.08889.json

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

[![Syntology run record](https://syntology.ai/api/ran/1904.08889.svg)](https://syntology.ai/paper/1904.08889)

Connect an agent — have() is free.