SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2205.00705 · ECCV · 2022

3D Object Detection with a Self-supervised Lidar Scene Flow Backbone

Alois Knoll, Mingyu Liu, Emeç Erçelik, Ekim Yurtsever, Zhijie Yang, Hanzhen Zhang, Pınar Topçam, Maximilian Listl, Yılmaz Kaan C ¸aylı

arXiv · PDF · Open in the Atlas

Code that ran

10 of 16 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 10 have no contract verdict; 0 failures of our instrument; 6 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/2205.00705.json · badge: /api/ran/2205.00705.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
WeijingShi/Point-GNN canonical (classifier label)
classifier basis not recorded on the link
10 ran · 5 no recorded run
HimangiM/Just-Go-with-the-Flow-Self-Supervised-Scene-Flow-Estimation community
Papers with Code archive
0 ran · 1 no recorded run
FunctionStatusWhere it lives
box3d_to_cam_points Ran
ran on a synthesized input
WeijingShi/Point-GNN/dataset/kitti_dataset.py
code served (permissive licence) · get_code("0141f393c163c9c5")
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 Points. 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
from collections import namedtuple

Points = namedtuple('Points', ['xyz', 'attr'])

# Construct a valid label dictionary with all required keys
label = {
    'x3d': 2.5,           # x position in camera coordinates
    'y3d': 1.2,           # y position in camera coordinates
    'z3d': 15.0,          # z position in camera coordinates
    'yaw': 0.5,           # rotation angle in radians
    'height': 1.5,        # height of the 3D box
    'width': 1.8,         # width of the 3D box
    'length': 4.0,        # length of the 3D box
}

# Define expand_factor with distinct values for each dimension
expend_factor = (1.1, 1.2, 1.3)

ARGS = {
    'label': label,
    'expend_factor': expend_factor,
}
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 Points — a Points 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
boxes_3d_to_corners Ran
ran on a synthesized input
WeijingShi/Point-GNN/models/nms.py
code served (permissive licence) · get_code("7283711d920302c1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family float, shape (2, 8, 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

# boxes_3d_to_corners expects a list/array of boxes in 3D
# Each box is represented as: [x3d, y3d, z3d, l, h, w, yaw]
# where x3d, y3d, z3d are center coordinates
# l is length, h is height, w is width
# yaw is rotation angle around y-axis

boxes_3d = np.array([
    [0.0, 0.0, 0.0, 2.0, 1.5, 1.0, 0.0],  # box 1: centered at origin, no rotation
    [1.0, -0.5, 2.0, 1.5, 1.2, 0.8, np.pi/4],  # box 2: offset position, 45 degree rotation
])

ARGS = {
    "boxes_3d": boxes_3d
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (2, 8, 3) — shape (2, 8, 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
center_box_encoding Ran
ran on a synthesized input
WeijingShi/Point-GNN/models/box_encoding.py
code served (permissive licence) · get_code("4855767c4a56c8c8")
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 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create small tensors for the function
# cls_labels: class labels for each point
cls_labels = torch.tensor([0, 1])

# points_xyz: 3D coordinates of points (batch_size=2, 3 coordinates)
points_xyz = torch.tensor([
    [1.0, 2.0, 3.0],
    [4.0, 5.0, 6.0]
], dtype=torch.float32)

# boxes_3d: 3D box parameters (batch_size=2, some number of box params)
# The function modifies the first 3 columns (center x, y, z)
boxes_3d = torch.tensor([
    [10.0, 20.0, 30.0, 5.0],
    [40.0, 50.0, 60.0, 8.0]
], dtype=torch.float32)

ARGS = {
    "cls_labels": cls_labels,
    "points_xyz": points_xyz,
    "boxes_3d": boxes_3d
}
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 2, dtype family float, shape (2, 4) — shape (2, 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
direct_box_decoding Ran
ran on a synthesized input
WeijingShi/Point-GNN/models/box_encoding.py
code served (permissive licence) · get_code("be80ed9b38980068")
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 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Build minimal valid arguments for direct_box_decoding
# The function takes cls_labels, points_xyz, and encoded_boxes
# and returns encoded_boxes unchanged

cls_labels = torch.tensor([0, 1], dtype=torch.long)  # shape (2,)

points_xyz = torch.randn(2, 3)  # shape (2, 3) - 2 points with xyz coordinates

encoded_boxes = torch.randn(2, 4)  # shape (2, 4) - 2 boxes with 4 encoding values

ARGS = {
    "cls_labels": cls_labels,
    "points_xyz": points_xyz,
    "encoded_boxes": encoded_boxes,
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
direct_box_encoding Ran
ran on a synthesized input
WeijingShi/Point-GNN/models/box_encoding.py
code served (permissive licence) · get_code("9e3ebce01f03bd3d")
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, 7) 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)

# The function simply returns boxes_3d, so we need to construct valid arguments
# that match the expected types based on typical 3D object detection conventions

# cls_labels: class labels, typically a tensor or list of class indices
cls_labels = torch.tensor([0, 1], dtype=torch.long)

# points_xyz: 3D points, typically shape (batch, num_points, 3) or (num_points, 3)
points_xyz = torch.randn(2, 8, 3, dtype=torch.float32)

# boxes_3d: 3D bounding boxes, typically shape (batch, num_boxes, 7) for (x, y, z, length, width, height, rotation)
# or similar representation
boxes_3d = torch.randn(2, 4, 7, dtype=torch.float32)

ARGS = {
    "cls_labels": cls_labels,
    "points_xyz": points_xyz,
    "boxes_3d": boxes_3d,
}
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 Tensor, ndim 3, dtype family float, shape (2, 4, 7) — shape (2, 4, 7) 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
downsample_by_average_voxel Ran
ran on a synthesized input
WeijingShi/Point-GNN/dataset/kitti_dataset.py
code served (permissive licence) · get_code("ee3750781dc56c2b")
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 Points. 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
from collections import namedtuple

Points = namedtuple('Points', ['xyz', 'attr'])

# Create small sample point cloud data
np.random.seed(0)

# xyz: N points with 3D coordinates
# Create 8 points in a small space to allow for meaningful voxel downsampling
xyz = np.array([
    [0.1, 0.2, 0.3],
    [0.15, 0.25, 0.35],
    [1.5, 1.6, 1.7],
    [1.55, 1.65, 1.75],
    [3.0, 3.1, 3.2],
    [3.05, 3.15, 3.25],
    [5.0, 5.1, 5.2],
    [5.05, 5.15, 5.25],
], dtype=np.float32)

# attr: attributes for each point (e.g., RGB values or other properties)
# Using 4 values per point to have distinct dimension
attr = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [1.1, 2.1, 3.1, 4.1],
    [5.0, 6.0, 7.0, 8.0],
    [5.1, 6.1, 7.1, 8.1],
    [10.0, 11.0, 12.0, 13.0],
    [10.1, 11.1, 12.1, 13.1],
    [15.0, 16.0, 17.0, 18.0],
    [15.1, 16.1, 17.1, 18.1],
], dtype=np.float32)

points = Points(xyz=xyz, attr=attr)

# voxel_size: scalar defining the size of voxel cells
voxel_size = 1.0

ARGS = {
    'points': points,
    'voxel_size': voxel_size,
}
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 Points — a Points 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
downsample_by_random_voxel Ran
ran on a synthesized input
WeijingShi/Point-GNN/dataset/kitti_dataset.py
code served (permissive licence) · get_code("384b9e6949fc0b72")
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 Points. 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 random
from collections import namedtuple

Points = namedtuple('Points', ['xyz', 'attr'])

# Create sample point cloud data
num_points = 10
xyz = np.random.uniform(0, 100, size=(num_points, 3)).astype(np.float32)
attr = np.random.uniform(0, 1, size=(num_points, 4)).astype(np.float32)
points = Points(xyz=xyz, attr=attr)

voxel_size = 5.0

ARGS = {
    'points': points,
    'voxel_size': voxel_size,
    'add_rnd3d': False
}
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 Points — a Points 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
load_config Ran
ran on a synthesized input
WeijingShi/Point-GNN/util/config_util.py
code served (permissive licence) · get_code("434c5a9cd34a6dcc")
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 dict. 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 json
import tempfile
import os

# Create a temporary configuration file
temp_dir = tempfile.mkdtemp()
config_file = os.path.join(temp_dir, "config.json")

# Write a valid JSON configuration
config_data = {
    "model": "test_model",
    "learning_rate": 0.001,
    "batch_size": 32,
    "epochs": 10,
    "layers": [64, 128, 256]
}

with open(config_file, 'w') as f:
    json.dump(config_data, f)

ARGS = {
    "filename": config_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
load_train_config Ran
ran on a synthesized input
WeijingShi/Point-GNN/util/config_util.py
code served (permissive licence) · get_code("2df98c5f04b6fb17")
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 dict. 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 json
import tempfile
import os

# Create a temporary config file with valid JSON content
temp_dir = tempfile.gettempdir()
config_file = os.path.join(temp_dir, 'test_train_config.json')

# Write a sample config dictionary to the file
sample_config = {
    "learning_rate": 0.001,
    "batch_size": 32,
    "epochs": 10,
    "model_type": "transformer",
    "hidden_dim": 256
}

with open(config_file, 'w') as f:
    json.dump(sample_config, f)

ARGS = {
    "filename": config_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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_layer_downsampling_random Ran
ran on a synthesized input
WeijingShi/Point-GNN/models/graph_gen.py
code served (permissive licence) · get_code("5794795650533553")
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
import random

random.seed(0)
np.random.seed(0)

# Create a small point cloud with 5 points in 3D space
points_xyz = np.array([
    [0.0, 0.0, 0.0],
    [0.5, 0.5, 0.5],
    [1.0, 1.0, 1.0],
    [1.5, 1.5, 1.5],
    [2.0, 2.0, 2.0]
], dtype=np.float32)

base_voxel_size = 0.5

levels = [1, 2]

add_rnd3d = False

ARGS = {
    "points_xyz": points_xyz,
    "base_voxel_size": base_voxel_size,
    "levels": levels,
    "add_rnd3d": add_rnd3d
}
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
average_gradients Not yet run
unverified (harvested, no recorded run)
WeijingShi/Point-GNN/util/tf_util.py
code served (permissive licence) · get_code("a0172c48afaa4a94")
How it ran

no invocation recorded.

focal_loss_sigmoid Not yet run
unverified (harvested, no recorded run)
WeijingShi/Point-GNN/models/loss.py
code served (permissive licence) · get_code("dce4ab6cd7b4540f")
How it ran

no invocation recorded.

focal_loss_softmax Not yet run
unverified (harvested, no recorded run)
WeijingShi/Point-GNN/models/loss.py
code served (permissive licence) · get_code("9fc528f84f0af21b")
How it ran

no invocation recorded.

get_model_flow Not yet run
unverified (harvested, no recorded run)
HimangiM/Just-Go-with-the-Flow-Self-Supervised-Scene-Flow-Estimation/src/model_concat_upsa_cycle.py
code served (permissive licence) · get_code("e1762f3d022bfacd")
How it ran

no invocation recorded.

instance_normalization Not yet run
unverified (harvested, no recorded run)
WeijingShi/Point-GNN/models/gnn.py
code served (permissive licence) · get_code("cbc514a51f3ff453")
How it ran

no invocation recorded.

overlapped_boxes_3d Not yet run
unverified (harvested, no recorded run)
WeijingShi/Point-GNN/models/nms.py
code served (permissive licence) · get_code("487eda418ebb9106")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 16 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.

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

State-of-the-art lidar-based 3D object detection methods rely on supervised learning and large labeled datasets. However, annotating lidar data is resource-consuming, and depending only on supervised learning limits the applicability of trained models. Self-supervised training strategies can alleviate these issues by learning a general point cloud backbone model for downstream 3D vision tasks. Against this backdrop, we show the relationship between self-supervised multi-frame flow representations and single-frame 3D detection hypotheses. Our main contribution leverages learned flow and motion representations and combines a self-supervised backbone with a supervised 3D detection head. First, a self-supervised scene flow estimation model is trained with cycle consistency. Then, the point cloud encoder of this model is used as the backbone of a single-frame 3D object detection head model. This second 3D object detection model learns to utilize motion representations to distinguish dynamic objects exhibiting different movement patterns. Experiments on KITTI and nuScenes benchmarks show that the proposed self-supervised pre-training increases 3D detection performance significantly. https://github.com/emecercelik/ssl-3d-detection.git

For agents

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

get_harvested_code_for_paper("2205.00705")
get_code_for_paper("2205.00705")
have("2205.00705")

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

curl https://syntology.ai/api/ran/2205.00705.json

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

[![Syntology run record](https://syntology.ai/api/ran/2205.00705.svg)](https://syntology.ai/paper/2205.00705)

Connect an agent — have() is free.