SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2207.04997 · ECCV · 2022

A Closer Look at Invariances in Self-supervised Pre-training for 3D Vision

Lanxiao Li

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 8 have no contract verdict; 0 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/2207.04997.json · badge: /api/ran/2207.04997.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
lilanxiao/invar3d community (classifier label)
classifier basis not recorded on the link
9 ran · 3 no recorded run
FunctionStatusWhere it lives
concat_all_gather Ran
ran, honoured the contract we drafted
lilanxiao/invar3d/loss_moco.py
pointer only (licence: MIT); the stored copy get_code serves, from yibingwei-1/latentmim, is pointer only (licence: NONE) · get_code("8cacb1e15bf1d910")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from yibingwei-1/latentmim, and GitHub detected no licence in that repository when it was checked. 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.

contrast_in_bacth Ran
ran on a synthesized input
lilanxiao/invar3d/loss.py
code served (permissive licence) · get_code("756e7cffdd4b7f0a")
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 torch

torch.manual_seed(0)

# Construct tiny tensors for f1 and f2
# Shape: (B, C, N) where B=batch, C=channels, N=points
B, C, N = 2, 4, 3
f1 = torch.randn(B, C, N)
f2 = torch.randn(B, C, N)

# Temperature parameter
t = 0.1

# Symmetric flag
symmetric = True

ARGS = {
    'f1': f1,
    'f2': f2,
    't': t,
    'symmetric': symmetric
}
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 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
contrast_in_scene Ran
ran on a synthesized input
lilanxiao/invar3d/loss.py
code served (permissive licence) · get_code("c9d418ddc1cfc9c2")
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 torch

torch.manual_seed(0)

# Small tensors: B=2 (batch), C=4 (channels), N=3 (points)
B, C, N = 2, 4, 3

f1 = torch.randn(B, C, N)
f2 = torch.randn(B, C, N)
t = 0.07
symmetric = True

ARGS = {
    'f1': f1,
    'f2': f2,
    't': t,
    'symmetric': symmetric
}
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 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
depth2points Ran
ran on a synthesized input
lilanxiao/invar3d/scannet/data_utils.py
code served (permissive licence) · get_code("ab15951e35e67a98")
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 2, dtype family float, shape (20, 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 a small depth map (H=4, W=5)
depth_map = np.random.rand(4, 5) * 5 + 0.5  # depth values between 0.5 and 5.5

# Create a camera intrinsic matrix K (3x3)
# K = [[fx,  0, cx],
#      [ 0, fy, cy],
#      [ 0,  0,  1]]
K = np.array([
    [500.0,   0.0, 2.5],    # fx, 0, cx
    [  0.0, 500.0, 2.0],    # 0, fy, cy
    [  0.0,   0.0, 1.0]     # 0, 0, 1
], dtype=np.float32)

ARGS = {
    'depth_map': depth_map,
    'K': K
}
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 2, dtype family float, shape (20, 3) — shape (20, 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_normalize_2d3d Ran
ran on a synthesized input
lilanxiao/invar3d/loss.py
code served (permissive licence) · get_code("4fe4c96dd9f15ede")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Analyze the function:
# - fm: torch.Tensor, used as fm.size() -> (B, C, H, W), then view to (B, C, -1)
# - fm_mask: torch.Tensor, used in _gather(fm_lin, fm_mask) -> shape (B, M)
# - seeds_features: torch.Tensor, used in _gather(seeds_features, ind3d) -> shape (B, C, N)
# - ind2d: torch.Tensor, used in _gather(fm_lin_sample, ind2d) -> shape (B, M2)
# - ind3d: torch.Tensor, used in _gather(seeds_features, ind3d) -> shape (B, M3)

B = 2  # batch size
C = 4  # channel dimension
H = 3  # height
W = 5  # width
N = 7  # number of seed features
M = 6  # mask sample count
M2 = 5  # ind2d sample count
M3 = 3  # ind3d sample count

# fm: (B, C, H, W)
fm = torch.randn(B, C, H, W)

# fm_mask: (B, M) - indices into the H*W = 15 spatial positions
fm_mask = torch.randint(0, H * W, (B, M))

# seeds_features: (B, C, N)
seeds_features = torch.randn(B, C, N)

# ind2d: (B, M2) - indices into the M = 6 positions from fm_lin_sample
ind2d = torch.randint(0, M, (B, M2))

# ind3d: (B, M3) - indices into the N = 7 positions of seeds_features
ind3d = torch.randint(0, N, (B, M3))

ARGS = {
    "fm": fm,
    "fm_mask": fm_mask,
    "seeds_features": seeds_features,
    "ind2d": ind2d,
    "ind3d": ind3d,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
global_features Ran
ran on a synthesized input
lilanxiao/invar3d/loss_moco.py
code served (permissive licence) · get_code("b74158904f710a22")
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
from torch import nn
from torch import Tensor

torch.manual_seed(0)

# Test with 3D tensor (B, C, L) - batch of 2, channels of 4, length of 3
x_3d = torch.randn(2, 4, 3)

# Test with 4D tensor (B, C, H, W) - batch of 2, channels of 4, height of 3, width of 5
x_4d = torch.randn(2, 4, 3, 5)

ARGS = {
    'x': x_3d,
    'pool': 'max'
}
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
read_scannet_instri Ran
ran on a synthesized input
lilanxiao/invar3d/scannet/data_utils.py
code served (permissive licence) · get_code("0d36479bd8145ff0")
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 2, dtype family float, shape (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 tempfile
import os

# Create a temporary file with ScanNet calibration data
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "intrinsics.txt")

# Write sample ScanNet intrinsics file content
intrinsics_content = """
some header line
m_calibrationDepthIntrinsic = 577.870605 0.000000 319.5 0.000000 577.870605 239.5 0.000000 0.000000 1.000000 0.0 0.0 0.0 0.0 0.0 0.0 0.0
another line
"""

with open(temp_file, 'w') as f:
    f.write(intrinsics_content)

ARGS = {
    'path': temp_file
}
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 2, dtype family float, shape (3, 3) — shape (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
remove_zeros Ran
ran on a synthesized input
lilanxiao/invar3d/scannet/data_utils.py
code served (permissive licence) · get_code("e9a8d842472af519")
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 2, dtype family float, shape (4, 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 a small point cloud with shape (5, 3)
# The function indexes [:, 1] so we need at least 2 columns
# Column 1 will be compared against threshold
point_cloud = np.array([
    [0.0, 0.05, 0.2],
    [0.1, 0.15, 0.3],
    [0.2, 0.25, 0.4],
    [0.3, 0.35, 0.5],
    [0.4, 0.45, 0.6]
], dtype=np.float32)

ARGS = {
    "point_cloud": point_cloud,
    "threshhold": 0.1,
    "return_mask": 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 ndarray, ndim 2, dtype family float, shape (4, 3) — shape (4, 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
revert_depth Ran
ran on a synthesized input
lilanxiao/invar3d/scannet/depth_histogram.py
code served (permissive licence) · get_code("b9c584cfcb20e22d")
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 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 numpy as np

# Create a small numpy array for testing
# The function treats this as depth values, so we'll create an array with some values
arr = np.array([
    [0.5, 1.2, 3.5, 0.002],
    [2.1, 0.015, 4.8, 1.1]
], dtype=np.float32)

max_depth = 8.0

ARGS = {
    "arr": arr,
    "max_depth": max_depth
}
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 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
batch_shuffle_ddp Not yet run
unverified (harvested, no recorded run)
lilanxiao/invar3d/loss_moco.py
code served (permissive licence) · get_code("ad25233af6c3d190")
How it ran

no invocation recorded.

get_split Not yet run
unverified (harvested, no recorded run)
lilanxiao/invar3d/scannet/sampler.py
code served (permissive licence) · get_code("a93412e4a677be68")
How it ran

no invocation recorded.

sample Not yet run
unverified (harvested, no recorded run)
lilanxiao/invar3d/scannet/sampler.py
code served (permissive licence) · get_code("4f5254c26b3f5bfb")
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, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 11 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
concat_all_gather 2 other papers
2 later
2407.15837 (2024-07)
Towards Latent Masked Image Modeling for Self-Supervised Visual Representation Learning
2503.13914 (2025-03)
PSA-SSL: Pose and Size-aware Self-Supervised Learning on LiDAR 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

Self-supervised pre-training for 3D vision has drawn increasing research interest in recent years. In order to learn informative representations, a lot of previous works exploit invariances of 3D features, e.g., perspective-invariance between views of the same scene, modalityinvariance between depth and RGB images, format-invariance between point clouds and voxels. Although they have achieved promising results, previous researches lack a systematic and fair comparison of these invariances. To address this issue, our work, for the first time, introduces a unified framework, under which various pre-training methods can be investigated. We conduct extensive experiments and provide a closer look at the contributions of different invariances in 3D pre-training. Also, we propose a simple but effective method that jointly pre-trains a 3D encoder and a depth map encoder using contrastive learning. Models pre-trained with our method gain significant performance boost in downstream tasks. For instance, a pre-trained VoteNet outperforms previous methods on SUN RGB-D and ScanNet object detection benchmarks with a clear margin.

For agents

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

get_harvested_code_for_paper("2207.04997")
get_code_for_paper("2207.04997")
have("2207.04997")

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

curl https://syntology.ai/api/ran/2207.04997.json

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

[![Syntology run record](https://syntology.ai/api/ran/2207.04997.svg)](https://syntology.ai/paper/2207.04997)

Connect an agent — have() is free.