SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2007.11576 · NeurIPS · 2020

Deep Variational Instance Segmentation

Li Fuxin, Jialin Yuan, Chao Chen

arXiv · PDF · Open in the Atlas

Code that ran

9 of 15 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (4 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 5 failures of our instrument; 6 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/2007.11576.json · badge: /api/ran/2007.11576.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
jia2lin3yuan1/2020-instanceSeg marked official in Papers with Code
Papers with Code archive
9 ran · 6 no recorded run
FunctionStatusWhere it lives
Concat Ran
ran on a synthesized input · constructed an object; no result was computed
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("aad3e10be1553162")
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 an instance of its own class, Concat: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# Create simple subnetworks that take input x and produce outputs
class SimpleNet(nn.Module):
    def __init__(self, input_dim, output_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x):
        return self.linear(x)

# Create two small networks with different output dimensions
net1 = SimpleNet(input_dim=4, output_dim=3)
net2 = SimpleNet(input_dim=4, output_dim=5)

nets = [net1, net2]

# extra_params is passed to torch.cat as **kwargs
# torch.cat accepts parameters like 'out' but the most common is none
extra_params = {}

ARGS = {
    'nets': nets,
    'extra_params': extra_params
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Concat — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Concat 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
InterpolateModule Ran
ran on a synthesized input · constructed an object; no result was computed
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("b33d21ad3788b76c")
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 an instance of its own class, InterpolateModule: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
import torch.nn.functional as F

torch.manual_seed(0)

# Create a small input tensor for interpolation
# Shape: (batch=2, channels=4, height=8, width=8)
input_tensor = torch.randn(2, 4, 8, 8)

# InterpolateModule takes *args and **kwdargs that get passed to F.interpolate
# F.interpolate signature: interpolate(input, size=None, scale_factor=None, mode='nearest', align_corners=None, ...)
# We'll set up arguments for upsampling by scale factor

ARGS = {
    'args': (),
    'kwdargs': {
        'size': (16, 16),
        'mode': 'bilinear',
        'align_corners': False
    }
}
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 InterpolateModule — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a InterpolateModule 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
RefineNet Ran
ran on a synthesized input · constructed an object; no result was computed
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("796be6419b89c41a")
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 an instance of its own class, RefineNet: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'roi_size': (14, 14),
    'fea_layers': [256],
    'num_classes': 21,
    'pi_margin': 1.0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type RefineNet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a RefineNet 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
RoIAlign Ran
ran on a synthesized input · constructed an object; no result was computed
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("f5376ed42c300495")
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 an instance of its own class, RoIAlign: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# Create input tensor: [batch_size=2, channels=4, height=16, width=16]
input_tensor = torch.randn(2, 4, 16, 16)

# Create bboxes tensor: [N, 5] with [batch_idx, x0, y0, x1, y1]
# Using N=3 bboxes across the batch
bboxes = torch.tensor([
    [0, 2.0, 2.0, 8.0, 8.0],    # batch 0
    [0, 4.0, 4.0, 10.0, 10.0],  # batch 0
    [1, 1.0, 1.0, 6.0, 6.0]     # batch 1
], dtype=torch.float32)

ARGS = {
    'out_size': (14, 14),
    'aligned': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type RoIAlign — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a RoIAlign 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
construct_backbone Ran
ran; our contract draft was wrong, not the code
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("b56ab4a1dcdcf54b")
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 MinimalBackbone. 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 torch.nn as nn

torch.manual_seed(0)

# Create a minimal backbone class that satisfies the function's requirements
class MinimalBackbone(nn.Module):
    def __init__(self, depth, width, in_channels=3):
        super().__init__()
        self.depth = depth
        self.width = width
        self.in_channels = in_channels
        self.layers = nn.ModuleList()
        # Initialize with some layers
        for i in range(depth):
            self.layers.append(nn.Linear(width, width))
    
    def add_layer(self):
        """Add a new layer to the backbone"""
        self.layers.append(nn.Linear(self.width, self.width))

# Create a minimal config class that has the required attributes
class BackboneConfig:
    def __init__(self):
        self.type = MinimalBackbone
        self.args = (2, 4)  # depth=2, width=4
        self.selected_layers = [0, 1, 2]  # max will be 2, so num_layers=3

cfg = BackboneConfig()

ARGS = {
    "cfg": cfg,
    "in_channels": 3
}
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 MinimalBackbone — a MinimalBackbone output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MinimalBackbone 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
extend_bboxes Ran
ran; our contract draft was wrong, not the code
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("41335d8912fa97b6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family int, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create sample bboxes tensor of shape [N, 4] with format [x0, y0, x1, y1]
# Using N=2 for batch size
bboxes = torch.tensor([
    [10.0, 20.0, 50.0, 80.0],
    [30.0, 40.0, 100.0, 120.0]
], dtype=torch.float32)

# Image height and width
ht = 480
wd = 640

# Scale parameter for expansion
scale = 0.2

# Maximum extension limits
max_ext = [16, 16]

# Minimum box size
minbox = [8, 8]

# Keep height-width ratio flag
keep_hw_ratio = False

ARGS = {
    "bboxes": bboxes,
    "ht": ht,
    "wd": wd,
    "scale": scale,
    "max_ext": max_ext,
    "minbox": minbox,
    "keep_hw_ratio": keep_hw_ratio
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family int, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_candidates Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("677fc7f6be074bda")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 torch.nn.functional as F

torch.manual_seed(0)

# Set up small tensors for testing
bs = 2  # batch size
ht = 4  # height
wd = 3  # width
total_pixels = ht * wd  # 12 pixels

# labelImgs: tensor in size [bs, ht*wd]
# Contains connected component labels (0, 1, 2, ...)
labelImgs = torch.tensor([
    [0, 0, 1, 1, 1, 2, 2, 0, 0, 3, 3, 3],
    [0, 1, 1, 2, 2, 2, 0, 0, 3, 3, 0, 0]
], dtype=torch.int64)

# intensityImgs: tensor in size [bs, 1, ht*wd]
# Intensity values for each pixel
intensityImgs = torch.tensor([
    [[0.5, 0.6, 0.7, 0.8, 0.75, 0.3, 0.35, 0.2, 0.15, 0.9, 0.85, 0.88]],
    [[0.4, 0.65, 0.7, 0.32, 0.33, 0.34, 0.1, 0.12, 0.91, 0.92, 0.2, 0.22]]
], dtype=torch.float32)

# xAxisI: tensor in size [1, ht*wd]
# x-coordinates for each pixel (column indices)
xAxisI = torch.tensor([[0, 1, 2, 0, 1, 2, 0, 1, 2, 0, 1, 2]], dtype=torch.float32)

# yAxisI: tensor in size [1, ht*wd]
# y-coordinates for each pixel (row indices)
yAxisI = torch.tensor([[0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3]], dtype=torch.float32)

# cand_params: parameters for filtering candidates
cand_params = {
    'bg_thr': 0.2,      # background threshold
    'size_thr': 2,      # size threshold
    'num_keep': 5       # number of candidates to keep
}

# merge_thr: intensity threshold for merging disconnected components
merge_thr = 0.15

ARGS = {
    'labelImgs': labelImgs,
    'intensityImgs': intensityImgs,
    'xAxisI': xAxisI,
    'yAxisI': yAxisI,
    'ht': ht,
    'wd': wd,
    'cand_params': cand_params,
    'merge_thr': merge_thr
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchvision — 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
extract_candidates_np Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("9045301263665532")
How it ran

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

import torch
import numpy as np
from skimage import measure as skmeasure

torch.manual_seed(0)

# Create a small binary label image (4x3)
labelI = np.array([
    [0, 1, 1],
    [0, 1, 1],
    [2, 2, 0],
    [2, 2, 0]
], dtype=np.uint8)

# Create a corresponding intensity image (4x3)
intensityI = np.array([
    [0.1, 0.5, 0.6],
    [0.2, 0.7, 0.8],
    [0.4, 0.5, 0.3],
    [0.6, 0.7, 0.2]
], dtype=np.float32)

# Batch index
bidx = 0

# Candidate parameters dictionary
cand_params = {
    'bg_thr': 0.2,
    'size_thr': 5,
    'num_keep': 100,
    'a2p_ratio_thr': 0.1,
    'a2p_pm_thr': 5.0
}

ARGS = {
    'labelI': labelI,
    'intensityI': intensityI,
    'bidx': bidx,
    'cand_params': cand_params
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, skimage, torch, torchvision — 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
make_net Ran
ran; our contract draft was wrong, not the code
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("82262a77afdbc139")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# Simple configuration for make_net
# Each layer is either:
# - [num_channels, kernel_size, kwargs_dict] for Conv2d/ConvTranspose2d/Interpolate
# - ['cat', [list of configs], kwargs_dict] for concatenation

conf = [
    [8, 3, {}],      # Conv2d: in_channels=3 -> out_channels=8, kernel_size=3
    [16, 3, {}],     # Conv2d: in_channels=8 -> out_channels=16, kernel_size=3
]

in_channels = 3
include_last_relu = True

ARGS = {
    "in_channels": in_channels,
    "conf": conf,
    "include_last_relu": include_last_relu,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchvision — 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
DVIS Not yet run
unverified (harvested, no recorded run)
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("17396a6efa6b88ea")
How it ran

no invocation recorded.

Discritizer Not yet run
unverified (harvested, no recorded run)
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("8bcc6fb31366b3c1")
How it ran

no invocation recorded.

FPN Not yet run
unverified (harvested, no recorded run)
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("7122b65e7ae3feed")
How it ran

no invocation recorded.

MeanshiftCluster_2 Not yet run
unverified (harvested, no recorded run)
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("b2ff3216632ff7c6")
How it ran

no invocation recorded.

RoIExtractor Not yet run
unverified (harvested, no recorded run)
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("e9da3ee1140401be")
How it ran

no invocation recorded.

vis_meanshift_result Not yet run
unverified (harvested, no recorded run)
jia2lin3yuan1/2020-instanceSeg/dvis_network.py
code served (permissive licence) · get_code("f0114052c90f2b45")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 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 15 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

Instance segmentation, which seeks to obtain both class and instance labels for each pixel in the input image, is a challenging task in computer vision. State-ofthe-art algorithms often employ a search-based strategy, which first divides the output image with a regular grid and generate proposals at each grid cell, then the proposals are classified and boundaries refined. In this paper, we propose a novel algorithm that directly utilizes a fully convolutional network (FCN) to predict instance labels. Specifically, we propose a variational relaxation of instance segmentation as minimizing an optimization functional for a piecewise-constant segmentation problem, which can be used to train an FCN end-to-end. It extends the classical Mumford-Shah variational segmentation algorithm to be able to handle the permutation-invariant ground truth in instance segmentation. Experiments on PASCAL VOC 2012 and the MSCOCO 2017 dataset show that the proposed approach efficiently tackles the instance segmentation task. The source code and trained models are released at https://github.com/jia2lin3yuan1/2020-instanceSeg.

For agents

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

get_harvested_code_for_paper("2007.11576")
get_code_for_paper("2007.11576")
have("2007.11576")

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

curl https://syntology.ai/api/ran/2007.11576.json

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

[![Syntology run record](https://syntology.ai/api/ran/2007.11576.svg)](https://syntology.ai/paper/2007.11576)

Connect an agent — have() is free.