SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2609.14722 · September 2026

PC 2 -AD: Point Cloud Upsampling to Safeguard 3D Anomaly Detection with Resolution-constrained Edge Devices

Jinbao Wang, Linlin Shen, Hanzhe Liang, Yingxi Xie, Jian Ning, Yutong Gu, Kejin Huang

arXiv · PDF · Open in the Atlas

Code that ran

25 of 30 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; 2 honoured the contract we drafted, 4 violated it and 0 have no contract verdict; 19 failures of our instrument; 5 no recorded run), measured 2026-11-22; 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/2609.14722.json · badge: /api/ran/2609.14722.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
gyutong406-commits/PC2-AD canonical (classifier label)
classifier basis: census
25 ran · 5 no recorded run
FunctionStatusWhere it lives
align_gt_to_test_order Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/common/gt_alignment.py
code served (permissive licence) · get_code("011a5646655b348d")
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

# Create small valid test data
# 4 points in 3D space
np.random.seed(0)
test_xyz = np.random.randn(4, 3).astype(np.float64)
gt_xyz = test_xyz.copy() + np.random.randn(4, 3) * 1e-7  # Near-identical with tiny perturbation
gt_labels = np.array([0, 1, 1, 0], dtype=np.uint8)  # Binary labels with at least one positive

ARGS = {
    'test_xyz': test_xyz,
    'gt_xyz': gt_xyz,
    'gt_labels': gt_labels,
    'tolerance': 1e-5,
    'n_jobs': 1,
    'require_positive': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, sklearn — 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
anomaly_shapenet_resolution Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/common/plan_schema.py
code served (permissive licence) · get_code("135ced43d7159bd2")
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 math
from typing import Any, Dict, Iterable, List, Mapping, Sequence, Tuple

# Construct valid arguments for anomaly_shapenet_resolution
train_point_counts = [100, 200, 150]

ARGS = {
    'train_point_counts': train_point_counts
}
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
audit_sample Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/common/audit_core.py
code served (permissive licence) · get_code("3f318d1063f508bf")
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 numpy as np
from pathlib import Path
import tempfile
import json

# Create a temporary directory for test files
temp_dir = tempfile.mkdtemp()
temp_path = Path(temp_dir)

# Create minimal test files
test_input_file = temp_path / "test_input.pcd"
test_input_file.write_text("# .PCD v.7 - Point Cloud Data file format\nVERSION .7\nFIELDS x y z\nSIZE 4 4 4\nTYPE f f f\nCOUNT 1 1 1\nWIDTH 2\nHEIGHT 1\nVIEWPOINT 0 0 0 1 0 0 0\nPOINTS 2\nDATA ascii\n0.0 0.0 0.0\n1.0 1.0 1.0\n")

gt_projected_file = temp_path / "gt_projected.npy"
np.save(gt_projected_file, np.array([1, 1], dtype=np.int64))

fps_indices_file = temp_path / "fps_indices.npy"
np.save(fps_indices_file, np.array([0, 1], dtype=np.int64))

source_index_file = temp_path / "source_index.npy"
np.save(source_index_file, np.array([0, 1], dtype=np.int64))

is_input_point_file = temp_path / "is_input_point.npy"
np.save(is_input_point_file, np.array([1, 1], dtype=np.int64))

round_id_file = temp_path / "round_id.npy"
np.save(round_id_file, np.array([0, 0], dtype=np.int64))

metadata_file = temp_path / "metadata.json"
metadata = {
    "dataset": "test_dataset",
    "subset": "test",
    "class_name": "test_class",
    "sample_name": "test_sample",
    "original_points": 2,
    "input_points": 2,
    "output_target_points": 2,
    "generation_rounds": 1,
    "gt_positive_count": 2
}
with open(metadata_file, 'w') as f:
    json.dump(metadata, f)

# Create a mock row dictionary
row = {
    'dataset': 'test_dataset',
    'subset': 'test',
    'class_name': 'test_class',
    'sample_name': 'test_sample',
    'original_points': 2,
    'input_points': 2,
    'output_target_points': 2,
    'generation_rounds': 1,
    'test_input_path': str(test_input_file),
    'gt_projected_path': str(gt_projected_file),
    'fps_indices_path': str(fps_indices_file),
    'source_index_path': str(source_index_file),
    'is_input_point_path': str(is_input_point_file),
    'round_id_path': str(round_id_file),
    'metadata_path': str(metadata_file),
    'source_test_path': str(test_input_file),
    'source_gt_path': '',
    'sample_type': 'normal',
    'gt_format': 'comma_xyz_label',
    'gt_match_tolerance': 0.1
}

# Create recomputed_fps_indices as optional parameter
recomputed_fps_indices = np.array([0, 1], dtype=np.int64)

ARGS = {
    'row': row,
    'recomputed_fps_indices': recomputed_fps_indices
}
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 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
augment_batch Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gyutong406-commits/PC2-AD/approach/train/contrastive_normality/contrastive_core.py
code served (permissive licence) · get_code("8dc4d570e3d816db")
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, 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 argparse
import torch
import math

torch.manual_seed(0)

# Create a small batch of point cloud data: [batch_size=2, num_points=4, coords=3]
points = torch.randn(2, 4, 3, dtype=torch.float32)

# Create argparse.Namespace with all required arguments
args = argparse.Namespace(
    rotate_z=True,
    scale_min=0.8,
    scale_max=1.2,
    jitter_std=0.01,
    jitter_clip=0.05,
    dropout_max=0.2
)

ARGS = {
    "points": points,
    "args": args
}
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, 3) — shape (2, 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
build_morton_blocks Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gyutong406-commits/PC2-AD/approach/data_preparation/common/fps_sampling.py
code served (permissive licence) · get_code("5021029aef1d8f78")
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 list. 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

# Create small test data
np.random.seed(0)

# Generate 7 random 3D points in the range [0, 10]
points = np.random.uniform(0, 10, size=(7, 3)).astype(np.float64)

# Set parameters
block_size = 3
bits = 8

# Build ARGS dict with all parameters for build_morton_blocks
ARGS = {
    "points": points,
    "block_size": block_size,
    "bits": bits,
}
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 list — a list 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
build_plan_rows Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/anomaly_shapenet/build_plan.py
code served (permissive licence) · get_code("baf2acbb57d1fef3")
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 list, ndim 1, 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 tempfile
from pathlib import Path
from unittest.mock import MagicMock, patch

# Create temporary directories for the test structure
temp_dir = tempfile.TemporaryDirectory()
dataset_root = Path(temp_dir.name) / "dataset"
output_root = Path(temp_dir.name) / "output"

dataset_root.mkdir(parents=True, exist_ok=True)
output_root.mkdir(parents=True, exist_ok=True)

# Create a mock class structure with required directories and files
class_dir = dataset_root / "test_class"
class_dir.mkdir(parents=True, exist_ok=True)

train_dir = class_dir / "train"
train_dir.mkdir(parents=True, exist_ok=True)

test_dir = class_dir / "test"
test_dir.mkdir(parents=True, exist_ok=True)

gt_dir = class_dir / "GT"
gt_dir.mkdir(parents=True, exist_ok=True)

# Create dummy PCD files
train_file = train_dir / "train_sample.pcd"
train_file.write_text("# dummy pcd\n")

test_file = test_dir / "test_positive.pcd"
test_file.write_text("# dummy pcd\n")

gt_file = gt_dir / "test_positive.txt"
gt_file.write_text("0,0,0,0\n")

# Mock the helper functions that are called but not defined in the snippet
def mock_read_pcd_point_count(path):
    return 1000

def mock_anomaly_shapenet_resolution(train_counts):
    return {
        'train_sample_count': len(train_counts),
        'train_point_sum': sum(train_counts),
        'train_mean_points': sum(train_counts) // len(train_counts) if train_counts else 0,
        'input_points': 2048,
        'input_rule': 'farthest_point_sample',
        'output_target_points': 2048,
        'output_rule': 'farthest_point_sample',
        'generation_rounds': 1,
    }

def mock_validate_plan_rows(rows, require_source_files=True):
    return rows

PLAN_VERSION = 1

# Patch the global functions
import sys
current_module = sys.modules[__name__]
current_module.read_pcd_point_count = mock_read_pcd_point_count
current_module.anomaly_shapenet_resolution = mock_anomaly_shapenet_resolution
current_module.validate_plan_rows = mock_validate_plan_rows
current_module.PLAN_VERSION = PLAN_VERSION

ARGS = {
    'dataset_root': dataset_root,
    'output_root': output_root,
    'subset': 'pcd',
    'fps_seed': 42,
    'fps_block_size': 4096,
    'fps_batch_size': 16,
    'morton_bits': 21,
    'gt_match_tolerance': 1e-06,
}
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 list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
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
build_state_arrays Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gyutong406-commits/PC2-AD/approach/data_preparation/common/output_writer.py
code served (permissive licence) · get_code("46807b270ff8372c")
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, ndim 2, dtype family int; 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 numpy as np

# Create a simple 1D array of indices for fps_indices
fps_indices = np.array([0, 5, 10, 15, 20], dtype=np.int64)

ARGS = {
    'fps_indices': fps_indices
}
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, ndim 2, dtype family int — a tuple output has no array shape of its own; none is stated
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
check_equal Ran
ran, violated the contract we drafted
gyutong406-commits/PC2-AD/environments/gradpu118fix/verify.py
code served (permissive licence) · get_code("a62fb7b850f2bdd8")
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 bool, ndim 0, dtype family float, shape () 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)

ARGS = {
    'name': 'test_equality_check',
    'actual': 'hello world',
    'expected': 'hello world'
}
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 bool, ndim 0, dtype family float, shape () — shape () 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
contains_forbidden_public_token Ran
ran, violated the contract we drafted
gyutong406-commits/PC2-AD/approach/data_preparation/common/plan_schema.py
code served (permissive licence) · get_code("25924b38d2bca762")
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 bool, ndim 0, dtype family float, shape () 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

ARGS = {
    "value": "test1"
}
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 bool, ndim 0, dtype family float, shape () — shape () 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
contrastive_loss_and_metrics Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/train/contrastive_normality/contrastive_core.py
code served (permissive licence) · get_code("cb133a02450c2bf7")
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, ndim 1, 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 torch.nn.functional as F

torch.manual_seed(0)

# Create small tensors for z1_raw and z2_raw
# The function expects:
# - z1_raw and z2_raw to have the same shape
# - batch size >= 2
# - They are normalized and used in matmul operations
# Using batch size 2, feature dimension 4

batch_size = 2
feature_dim = 4

z1_raw = torch.randn(batch_size, feature_dim)
z2_raw = torch.randn(batch_size, feature_dim)

temperature = 0.07

ARGS = {
    'z1_raw': z1_raw,
    'z2_raw': z2_raw,
    'temperature': temperature,
}
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, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
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
decode_h5_string Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/train/contrastive_normality/contrastive_core.py
code served (permissive licence) · get_code("809127532d86b9a4")
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 str. 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)

# The function decode_h5_string takes a single parameter 'x'
# It checks if x is bytes and decodes it as UTF-8, otherwise converts to string
# We'll test with a bytes object since that's the main use case

ARGS = {
    'x': b'hello world'
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
is_normal_sample Ran
ran, violated the contract we drafted
gyutong406-commits/PC2-AD/approach/data_preparation/anomaly_shapenet/build_plan.py
code served (permissive licence) · get_code("dd8f4aad21ca37dc")
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 bool, ndim 0, dtype family float, shape () 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 re

def is_normal_sample(name: str) -> bool:
    lower = name.lower()
    return 'positive' in lower or re.search('(^|_)good(?:_|$)', lower) is not None

ARGS = {
    "name": "positive_sample"
}
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 bool, ndim 0, dtype family float, shape () — shape () 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
load_config Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/normality_training/build_normality_data.py
code served (permissive licence) · get_code("bdd6242c037f6238")
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 tempfile
from pathlib import Path
import yaml

# Create a temporary directory and a valid YAML configuration file
temp_dir = tempfile.TemporaryDirectory()
config_path = Path(temp_dir.name) / "config.yaml"

# Write a valid YAML configuration
config_data = {
    "model": "test_model",
    "learning_rate": 0.001,
    "batch_size": 32,
    "epochs": 10,
    "nested": {
        "key1": "value1",
        "key2": 42
    }
}

with config_path.open('w', encoding='utf-8') as f:
    yaml.safe_dump(config_data, f)

ARGS = {
    "path": config_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — 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_gt_txt Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/common/gt_alignment.py
code served (permissive licence) · get_code("33ffd4e96640bc66")
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 tempfile
from pathlib import Path
import numpy as np

# Create a temporary file with valid ground truth data
with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt') as f:
    temp_path = f.name
    # Write some sample GT data: x, y, z, label, (and optional extra columns)
    f.write("1.0 2.0 3.0 1.0 5.0\n")
    f.write("4.0 5.0 6.0 0.0 7.0\n")
    f.write("7.0 8.0 9.0 1.0 8.0\n")

ARGS = {
    'path': Path(temp_path),
    'delimiter': None,
}
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
load_metadata Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/common/audit_core.py
code served (permissive licence) · get_code("795221c9231146f1")
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
from pathlib import Path

# Create a temporary JSON file with valid metadata
temp_dir = tempfile.TemporaryDirectory()
metadata_path = Path(temp_dir.name) / "metadata.json"

# Write a valid metadata dict to the file
metadata = {
    "version": "1.0",
    "name": "test_metadata",
    "description": "A test metadata file",
    "tags": ["test", "example"],
    "count": 42
}

with metadata_path.open('w', encoding='utf-8') as f:
    json.dump(metadata, f)

ARGS = {
    "path": metadata_path
}
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
morton_codes Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gyutong406-commits/PC2-AD/approach/data_preparation/common/fps_sampling.py
code served (permissive licence) · get_code("0476e3fd59c102cb")
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 int, 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 numpy as np

# Create valid arguments for morton_codes function
# points: np.ndarray with shape [N, 3] of float64 values
# bits: int in range [1, 21]

# Create a small set of 3D points
points = np.array([
    [0.1, 0.2, 0.3],
    [0.5, 0.6, 0.7],
    [0.9, 0.8, 0.4],
    [0.2, 0.3, 0.5],
], dtype=np.float64)

bits = 10

ARGS = {
    "points": points,
    "bits": bits,
}
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 int, 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
output_paths Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/anomaly_shapenet/build_plan.py
code served (permissive licence) · get_code("349f6f1344150971")
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.

from pathlib import Path
import tempfile

# Create a temporary directory to use as output_root
temp_dir = tempfile.mkdtemp()

ARGS = {
    'output_root': Path(temp_dir),
    'subset': 'train',
    'class_name': 'chair',
    'stem': 'sample_001'
}
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
parse_args Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/normality_training/build_normality_data.py
code served (permissive licence) · get_code("3b58384c53c034e6")
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 Namespace. 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 argparse
from typing import Optional, Sequence

def parse_args(argv: Optional[Sequence[str]]=None) -> argparse.Namespace:
    parser = argparse.ArgumentParser(description='Build normality-training data through one configuration-driven public entry.')
    parser.add_argument('--config', type=str, required=True)
    parser.add_argument('--max_samples', '--max-samples', dest='max_samples', type=int, default=0)
    parser.add_argument('--max_patches_per_sample', '--max-patches-per-sample', dest='max_patches_per_sample', type=int, default=0)
    parser.add_argument('--overwrite', action='store_true')
    parser.add_argument('--validate_only', '--validate-only', dest='validate_only', action='store_true')
    return parser.parse_args(argv)

# Setup code to construct valid arguments
ARGS = {
    'argv': [
        '--config', '/path/to/config.yaml',
        '--max_samples', '100',
        '--max_patches_per_sample', '10',
        '--overwrite',
        '--validate_only'
    ]
}
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 Namespace — a Namespace output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Namespace 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
parse_config_gpu_ids Ran
ran, honoured the contract we drafted
gyutong406-commits/PC2-AD/approach/data_preparation/normality_training/build_normality_data.py
code served (permissive licence) · get_code("cba24a83b92c41f9")
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 list, ndim 1, dtype family int; 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
from typing import Any, Dict, List, Mapping, Optional, Sequence

torch.manual_seed(0)

# parse_config_gpu_ids takes one parameter: value
# The function checks if value is a str or list, and processes accordingly
# Valid test cases should pass a str or list of integers

ARGS = {
    "value": "0,1,2"  # A valid comma-separated string of GPU IDs
}
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 list, ndim 1, dtype family int — a list output has no array shape of its own; none is stated
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_pcd_point_count Ran
ran, honoured the contract we drafted
gyutong406-commits/PC2-AD/approach/data_preparation/common/pcd_io.py
code served (permissive licence) · get_code("c8a6713868d17edc")
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 int, ndim 0, dtype family int, shape () 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
from pathlib import Path

# Create a temporary PCD file for testing
temp_dir = tempfile.TemporaryDirectory()
temp_path = Path(temp_dir.name) / "test.pcd"

# Write a minimal valid PCD file with POINTS field
pcd_content = b"""# .PCD v.7 - Point Cloud Data file format
VERSION .7
FIELDS x y z intensity
SIZE 4 4 4 4
TYPE f f f f
COUNT 1 1 1 1
WIDTH 100
HEIGHT 1
VIEWPOINT 0 0 0 1 0 0 0
POINTS 100
DATA ascii
0 0 0 0
"""

with temp_path.open('wb') as f:
    f.write(pcd_content)

ARGS = {
    'path': temp_path
}
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 int, ndim 0, dtype family int, shape () — shape () 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
real3d_ad_resolution Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/common/plan_schema.py
code served (permissive licence) · get_code("11363b99ceae2d1d")
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
import numpy as np

torch.manual_seed(0)

# Analyzing the function:
# - Takes original_points: int as parameter
# - Checks if original < 4, raises ValueError if so
# - Returns a dict with computed values based on original_points
# - No complex objects needed, just an integer

ARGS = {
    'original_points': 16
}
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 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
row_output_paths Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/anomaly_shapenet/prepare.py
code served (permissive licence) · get_code("89da1d0a91bf0395")
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
from pathlib import Path
from typing import Any, Mapping

# Construct a minimal row dict that satisfies the function's requirements
# The function expects a Mapping with string keys pointing to path-like values

row = {
    'test_input_path': '/tmp/test_input_0',
    'gt_projected_path': '/tmp/gt_projected_0',
    'fps_indices_path': '/tmp/fps_indices_0',
    'source_index_path': '/tmp/source_index_0',
    'is_input_point_path': '/tmp/is_input_point_0',
    'round_id_path': '/tmp/round_id_0',
    'metadata_path': '/tmp/metadata_0',
}

ARGS = {
    'row': row
}
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
split_by_3 Ran
ran, violated the contract we drafted
gyutong406-commits/PC2-AD/approach/data_preparation/common/fps_sampling.py
code served (permissive licence) · get_code("3ce2f4dbf0c5aa8f")
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 int, 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

# Create a small numpy array with 21-bit integer values
# The function expects values that fit in 21 bits (max value 2097151)
values = np.array([0, 1, 42, 2097151, 1234567], dtype=np.uint64)

ARGS = {
    "values": values
}
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 int, 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
validate_xyz Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/approach/data_preparation/common/pcd_io.py
code served (permissive licence) · get_code("11dd43439e267c4b")
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 valid points array with shape [N, 3] where N > 0
points = np.array([
    [1.0, 2.0, 3.0],
    [4.0, 5.0, 6.0],
    [7.0, 8.0, 9.0],
    [10.0, 11.0, 12.0]
], dtype=np.float32)

ARGS = {
    'points': points,
    'name': 'points'
}
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
version_of Ran
ran; our contract draft was wrong, not the code
gyutong406-commits/PC2-AD/environments/gradpu118fix/verify.py
code served (permissive licence) · get_code("91768e9a14df8e38")
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 str. 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 importlib

def version_of(module_name: str) -> str:
    module = importlib.import_module(module_name)
    return str(getattr(module, '__version__', 'installed'))

# The function takes a module_name string and imports it,
# then tries to get its __version__ attribute, defaulting to 'installed'
# We need to provide a valid module name that exists in the environment

ARGS = {
    'module_name': 'torch'
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
audit_plan Not yet run
unverified (harvested, no recorded run)
gyutong406-commits/PC2-AD/approach/data_preparation/anomaly_shapenet/audit.py
code served (permissive licence) · get_code("b3ac13ee3f64696c")
How it ran

no invocation recorded.

build_aligned_gt Not yet run
unverified (harvested, no recorded run)
gyutong406-commits/PC2-AD/approach/data_preparation/common/gt_alignment.py
code served (permissive licence) · get_code("9ec207cf13c2799c")
How it ran

no invocation recorded.

prepare_row Not yet run
unverified (harvested, no recorded run)
gyutong406-commits/PC2-AD/approach/data_preparation/anomaly_shapenet/prepare.py
code served (permissive licence) · get_code("b8d7a611007ed4b3")
How it ran

no invocation recorded.

read_pcd_points Not yet run
unverified (harvested, no recorded run)
gyutong406-commits/PC2-AD/approach/data_preparation/common/pcd_io.py
code served (permissive licence) · get_code("7f838db7f406a43e")
How it ran

no invocation recorded.

run_plan Not yet run
unverified (harvested, no recorded run)
gyutong406-commits/PC2-AD/approach/data_preparation/anomaly_shapenet/prepare.py
code served (permissive licence) · get_code("471844260d549674")
How it ran

no invocation recorded.

Also present in other papers' code

Of 30 samples lifted from this paper's linked repositories, 30 carry a function identity and 0 do not. Of the 30 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 30 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

Low-cost and low-resolution sensors used in edge deployments can produce test point clouds that are substantially sparser than the normal training data. This train-test sampling-resolution gap changes the local geometry available to a 3D anomaly detector. We propose PC$^2$-AD, a point cloud upsampling framework that compensates sparse test inputs before downstream detection. Target Domain Candidate Generation (TCG) adapts a pretrained upsampler to normal training geometry and generates a dense candidate pool. Geometry-Aware Candidate Filtering (GACF) selects candidates according to geometric spacing and spatial coverage. Normality-Preserving Point Compensation (NPPC) refines the selection by comparing candidate normality scores with those of their input anchors. The selected points are combined with the unchanged input points and processed by the existing detector. Experiments with six detectors on two Anomaly-ShapeNet settings and Real3D-AD show improvements in the mean of object-level and point-level AUROC for all six detectors in each Anomaly-ShapeNet setting and four on Real3D-AD. These results support point cloud compensation as an input-level approach to improving 3D anomaly detection under low-resolution sensing conditions. Code is publicly available at https://github.com/gyutong406-commits/PC2-AD.

For agents

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

get_harvested_code_for_paper("2609.14722")
get_code_for_paper("2609.14722")
have("2609.14722")

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

curl https://syntology.ai/api/ran/2609.14722.json

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

[![Syntology run record](https://syntology.ai/api/ran/2609.14722.svg)](https://syntology.ai/paper/2609.14722)

Connect an agent — have() is free.