SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2312.03015 · 2023

PartSLIP++: Enhancing Low-Shot 3D Part Segmentation via Multi-View Instance Segmentation and Maximum Likelihood Estimation

arXiv · PDF · Open in the Atlas

Code that ran

17 of 17 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; 8 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 8 failures of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2312.03015.json · badge: /api/ran/2312.03015.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
zyc00/partslip2 canonical (classifier label)
classifier basis: census
17 ran · 0 no recorded run
FunctionStatusWhere it lives
calc_iou Ran
ran, honoured the contract we drafted
zyc00/partslip2/eval/semantic_eval.py
code served (permissive licence) · get_code("7905eb68c83d6fc3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create binary prediction and ground truth arrays
# The function expects boolean or binary arrays that can be used with
# np.logical_and and np.logical_or

pred = np.array([
    [1, 0, 1, 0],
    [0, 1, 1, 0]
], dtype=bool)

gt = np.array([
    [1, 1, 0, 0],
    [0, 1, 0, 1]
], dtype=bool)

ARGS = {
    "pred": pred,
    "gt": gt
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_pc_within_bbox Ran
ran, violated the contract we drafted
zyc00/partslip2/src/bbox2seg.py
code served (permissive licence) · get_code("1b89b53acf1b369c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family float, shape (5,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create bounding box coordinates
x1 = 0.0
y1 = 0.0
x2 = 10.0
y2 = 10.0

# Create a point cloud with shape (N, 2) where N is the number of points
# Each row is [x, y] coordinates
pc = np.array([
    [1.0, 1.0],
    [5.0, 5.0],
    [15.0, 15.0],
    [-1.0, 5.0],
    [8.5, 9.5]
], dtype=np.float32)

ARGS = {
    'x1': x1,
    'y1': y1,
    'x2': x2,
    'y2': y2,
    'pc': pc
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (5,) — shape (5,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_pc_within_seg_mask Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zyc00/partslip2/src/bbox2seg.py
code served (permissive licence) · get_code("8a98fdde1d2d9243")
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 1, dtype family float, shape (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 torch
import numpy as np

torch.manual_seed(0)

# Create a segmentation mask [H, W]
# Using distinct dimensions: 5 height, 7 width
mask = torch.randint(0, 2, (5, 7), dtype=torch.float32)

# Create point cloud coordinates [N, 2]
# N=3 points, each with (x, y) screen coordinates
# x coordinates should be in range [0, 7), y coordinates in range [0, 5)
pc_coord = torch.tensor([
    [2, 1],
    [4, 3],
    [1, 2]
], dtype=torch.long)

ARGS = {
    "mask": mask,
    "pc_coord": pc_coord
}
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 1, dtype family float, shape (3,) — shape (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
compute_ap Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zyc00/partslip2/eval/eval_utils.py
code served (permissive licence) · get_code("ab81f972201a59c2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create small synthetic data for compute_ap function
# tp and fp are arrays of true positives and false positives
tp = np.array([1, 0, 1, 1, 0, 1, 0, 0, 1, 1], dtype=np.int32)
fp = np.array([0, 1, 0, 0, 1, 0, 1, 1, 0, 0], dtype=np.int32)

# gt_npos is the total number of ground truth positive instances
gt_npos = 7

# n_bins is the number of bins for precision-recall curve
n_bins = 100

# plot_fn is optional - None means no plotting
plot_fn = None

ARGS = {
    'tp': tp,
    'fp': fp,
    'gt_npos': gt_npos,
    'n_bins': n_bins,
    'plot_fn': plot_fn
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
draw_rectangle Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zyc00/partslip2/src/glip_inference.py
code served (permissive licence) · get_code("80ca23c0df5a69f0")
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 int, shape (100, 120, 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 sample image with shape (height, width, channels)
# Using dimensions that are distinct and large enough for the rectangle drawing
img = np.random.randint(0, 256, size=(100, 120, 3), dtype=np.uint8)

# Define rectangle coordinates
# x0, y0 = top-left corner
# x1, y1 = bottom-right corner
x0 = 20
y0 = 15
x1 = 80
y1 = 60

ARGS = {
    "img": img,
    "x0": x0,
    "y0": y0,
    "x1": x1,
    "y1": y1,
}
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 int, shape (100, 120, 3) — shape (100, 120, 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
gen_gt_labels_partnete Ran
ran, honoured the contract we drafted
zyc00/partslip2/eval/instance_eval.py
code served (permissive licence) · get_code("292627cba462fe52")
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 3, 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 os
import tempfile
import numpy as np
from pathlib import Path

# Create temporary directory structure with mock data files
temp_dir = tempfile.mkdtemp()
data_dir = Path(temp_dir) / 'data' / 'test' / 'category_a' / 'model_b'
data_dir.mkdir(parents=True, exist_ok=True)

# Create mock label data
sem_label = np.random.randint(0, 10, size=(32, 32), dtype=np.int32)
ins_label = np.random.randint(0, 20, size=(32, 32), dtype=np.int32)
label_dict = {
    'semantic_seg': sem_label,
    'instance_seg': ins_label
}

# Save the label file
label_path = data_dir / 'label.npy'
np.save(label_path, label_dict)

# Change to temp directory so relative path works
original_dir = os.getcwd()
os.chdir(temp_dir)

ARGS = {
    'category': 'category_a',
    'model': 'model_b'
}
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 3, 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
get_iou Ran
ran, honoured the contract we drafted
zyc00/partslip2/src/utils.py
code served (permissive licence) · get_code("743c650ee7589524")
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 float, 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)

# Construct two valid bounding boxes as dictionaries
# bb1: top-left at (10, 20), bottom-right at (50, 80)
bb1 = {
    'x1': 10.0,
    'y1': 20.0,
    'x2': 50.0,
    'y2': 80.0
}

# bb2: top-left at (30, 40), bottom-right at (70, 100)
# This overlaps with bb1
bb2 = {
    'x1': 30.0,
    'y1': 40.0,
    'x2': 70.0,
    'y2': 100.0
}

ARGS = {
    'bb1': bb1,
    'bb2': bb2
}
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 float, 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
get_point_iou Ran
ran, honoured the contract we drafted
zyc00/partslip2/src/bbox2seg.py
code served (permissive licence) · get_code("b21d7e74c2ec8b4c")
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 numpy as np

# Create two small binary point clouds (numpy arrays)
# The function uses np.logical_and, np.logical_or, and .sum()
# so we need boolean arrays

pc1 = np.array([
    [True, False, True, False],
    [False, True, False, True]
], dtype=bool)

pc2 = np.array([
    [True, True, False, False],
    [False, True, True, False]
], dtype=bool)

ARGS = {
    "pc1": pc1,
    "pc2": pc2
}
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 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
get_union Ran
ran, honoured the contract we drafted
zyc00/partslip2/src/mask2ins_refine.py
code served (permissive licence) · get_code("bf327ed3e34a1d26")
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 torch

# This function implements a union-find (disjoint set) data structure
# It takes a parent array `f` and an element `x`, and returns the root of the set containing `x`
# The function uses path compression

# f: a dictionary or list mapping each element to its parent
# x: an integer index into f

# Create a simple union-find structure
# This is a dictionary where f[i] points to the parent of element i
f = {0: 0, 1: 1, 2: 1, 3: 0}  # Element 0's parent is 0 (root), 1's parent is 1 (root), 2's parent is 1, 3's parent is 0

# x: the element we want to find the root of
x = 2

ARGS = {
    "f": f,
    "x": x
}
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 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
intersection Ran
ran; our contract draft was wrong, not the code
zyc00/partslip2/src/mask2ins_refine.py
code served (permissive licence) · get_code("db4d6d533b962001")
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

torch.manual_seed(0)

# The function takes two lists and finds their intersection
# lst1 and lst2 should be iterables that can be converted to sets

lst1 = [1, 2, 3, 4, 5]
lst2 = [3, 4, 5, 6, 7]

ARGS = {
    "lst1": lst1,
    "lst2": lst2
}
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
load_gt_h5 Ran
ran; our contract draft was wrong, not the code
zyc00/partslip2/eval/eval_utils.py
code served (permissive licence) · get_code("158388a7fdef7c78")
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 h5py
import tempfile
import numpy as np
import os

# Create a temporary HDF5 file with the required datasets
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test_gt.h5")

# Create sample data
B = 2  # batch size
N = 10  # number of points
K = 3  # number of masks

# Create the HDF5 file with required datasets
with h5py.File(temp_file, 'w') as f:
    # gt_mask: B x K x N, bool
    f.create_dataset('gt_mask', data=np.random.randint(0, 2, (B, K, N), dtype=bool))
    
    # gt_mask_label: B x K, uint8
    f.create_dataset('gt_mask_label', data=np.random.randint(0, 256, (B, K), dtype=np.uint8))
    
    # gt_mask_valid: B x K, bool
    f.create_dataset('gt_mask_valid', data=np.random.randint(0, 2, (B, K), dtype=bool))
    
    # gt_mask_other: B x N, bool
    f.create_dataset('gt_mask_other', data=np.random.randint(0, 2, (B, N), dtype=bool))

ARGS = {
    'fn': temp_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
h5py, 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_img Ran
ran, honoured the contract we drafted
zyc00/partslip2/src/glip_inference.py
code served (permissive licence) · get_code("4faf8454539ef598")
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 int, shape (4, 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
from PIL import Image
import numpy as np

# Create a temporary RGB image file
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test_image.png")

# Create a small RGB image (4x3 pixels with 3 channels)
img_array = np.random.randint(0, 256, (4, 3, 3), dtype=np.uint8)
pil_img = Image.fromarray(img_array, mode='RGB')
pil_img.save(temp_file)

ARGS = {
    'file_name': temp_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, 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 int, shape (4, 3, 3) — shape (4, 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
load_img Ran
ran, honoured the contract we drafted
zyc00/partslip2/src/mask2ins_refine.py
code served (permissive licence) · get_code("66f8aac6d7748b1b")
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 int, shape (4, 3, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from PIL import Image
import tempfile
import os

# Create a temporary RGB image file
temp_dir = tempfile.gettempdir()
temp_file = os.path.join(temp_dir, "test_image.png")

# Create a small RGB image (4x3 pixels)
rgb_array = np.random.randint(0, 256, (4, 3, 3), dtype=np.uint8)
pil_img = Image.fromarray(rgb_array, mode='RGB')
pil_img.save(temp_file)

ARGS = {
    'file_name': temp_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, 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 int, shape (4, 3, 3) — shape (4, 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
load_ply Ran
ran; our contract draft was wrong, not the code
zyc00/partslip2/eval/instance_eval.py
code served (permissive licence) · get_code("adc92675e71d2ebc")
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
import os

# Create a temporary PLY file with valid content
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.ply")

# Write a minimal valid PLY file
ply_content = """ply
format ascii 1.0
element vertex 3
property float x
property float y
property float z
property float red
property float green
property float blue
end_header
0.1 0.2 0.3 255.0 128.0 64.0
1.5 2.5 3.5 200.0 100.0 50.0
-1.0 -2.0 -3.0 50.0 150.0 200.0
"""

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

ARGS = {
    "file_name": 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 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_pred_h5 Ran
ran; our contract draft was wrong, not the code
zyc00/partslip2/eval/eval_utils.py
code served (permissive licence) · get_code("54d843892d13a52f")
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 h5py
import tempfile
import numpy as np

# Create a temporary HDF5 file with the required structure
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.h5')
temp_filename = temp_file.name
temp_file.close()

# Write test data to the HDF5 file
with h5py.File(temp_filename, 'w') as fout:
    # B=2 (batch), K=3 (number of parts), N=4 (mask dimension)
    B, K, N = 2, 3, 4
    
    # mask: B x K x N bool
    mask_data = np.random.rand(B, K, N) > 0.5
    fout.create_dataset('mask', data=mask_data)
    
    # label: B x K uint8
    label_data = np.random.randint(0, 10, size=(B, K), dtype=np.uint8)
    fout.create_dataset('label', data=label_data)
    
    # valid: B x K bool
    valid_data = np.random.rand(B, K) > 0.5
    fout.create_dataset('valid', data=valid_data)
    
    # conf: B x K float32
    conf_data = np.random.rand(B, K).astype(np.float32)
    fout.create_dataset('conf', data=conf_data)

ARGS = {
    'fn': temp_filename
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
h5py, 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
merge_small_components Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zyc00/partslip2/src/gen_superpoint.py
code served (permissive licence) · get_code("d188e48d8dedbdee")
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 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
from sklearn.neighbors import NearestNeighbors

# Create small XYZ point cloud data
np.random.seed(0)
xyz = np.random.randn(20, 3).astype(np.float32)

# Create components - list of lists containing indices into xyz
# Some components will be large (> min_component_size), some small
components = [
    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11],  # large component (12 points)
    [12, 13],  # small component (2 points)
    [14, 15, 16, 17, 18],  # small component (5 points)
    [19],  # small component (1 point)
]

ARGS = {
    "components": components,
    "xyz": xyz,
    "min_component_size": 10
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 2, 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
normalize_pc Ran
ran, honoured the contract we drafted
zyc00/partslip2/src/utils.py
code served (permissive licence) · get_code("e767dc9c06cc5761")
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 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
from unittest.mock import Mock

torch.manual_seed(0)

# Create a mock PointCloud object that has the methods used in normalize_pc
class MockPointCloud:
    def __init__(self):
        self.xyz = torch.randn(4, 3)
        self.rgb = torch.rand(4, 3)
    
    def points_padded(self):
        return self.xyz
    
    def features_padded(self):
        return self.rgb

# Create a mock io object with load_pointcloud method
class MockIO:
    def load_pointcloud(self, pc_file, device=None):
        return MockPointCloud()

ARGS = {
    'pc_file': '/tmp/test_pc.ply',
    'save_dir': '/tmp',
    'io': MockIO(),
    'device': 'cpu',
    'save_normalized_pc': False
}
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 tuple, ndim 3, 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

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 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 17 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

Open-world 3D part segmentation is pivotal in diverse applications such as robotics and AR/VR. Traditional supervised methods often grapple with limited 3D data availability and struggle to generalize to unseen object categories. PartSLIP, a recent advancement, has made significant strides in zero- and few-shot 3D part segmentation. This is achieved by harnessing the capabilities of the 2D open-vocabulary detection module, GLIP, and introducing a heuristic method for converting and lifting multi-view 2D bounding box predictions into 3D segmentation masks. In this paper, we introduce PartSLIP++, an enhanced version designed to overcome the limitations of its predecessor. Our approach incorporates two major improvements. First, we utilize a pre-trained 2D segmentation model, SAM, to produce pixel-wise 2D segmentations, yielding more precise and accurate annotations than the 2D bounding boxes used in PartSLIP. Second, PartSLIP++ replaces the heuristic 3D conversion process with an innovative modified Expectation-Maximization algorithm. This algorithm conceptualizes 3D instance segmentation as unobserved latent variables, and then iteratively refines them through an alternating process of 2D-3D matching and optimization with gradient descent. Through extensive evaluations, we show that PartSLIP++ demonstrates better performance over PartSLIP in both low-shot 3D semantic and instance-based object part segmentation tasks. Code released at https://github.com/zyc00/PartSLIP2.

For agents

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

get_harvested_code_for_paper("2312.03015")
get_code_for_paper("2312.03015")
have("2312.03015")

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

curl https://syntology.ai/api/ran/2312.03015.json

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

[![Syntology run record](https://syntology.ai/api/ran/2312.03015.svg)](https://syntology.ai/paper/2312.03015)

Connect an agent — have() is free.