SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.09829 · CVPR · 2024

Open-Vocabulary Semantic Segmentation with Image Embedding Balancing

Nong Sang, Changxin Gao, Yuanjie Shao, Dongyue Wu, Guilin Zhu, Xiangheng Shan, Adab Decoder

arXiv · PDF · Open in the Atlas

Code that ran

7 of 12 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 7 failures of our instrument; 5 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/2406.09829.json · badge: /api/ran/2406.09829.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
slonetime/EBSeg canonical (classifier label)
classifier basis: census
7 ran · 5 no recorded run
FunctionStatusWhere it lives
batch_dice_loss Ran
ran; our contract draft was wrong, not the code
slonetime/EBSeg/ebseg/model/matcher.py
code served (permissive licence) · get_code("bc2cb481a75c370d")
How it ran

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

import torch

torch.manual_seed(0)

# Create small tensors for inputs and targets
# inputs: arbitrary shape float tensor (predictions)
# targets: same shape as inputs, binary classification labels (0 or 1)
# Using shape (2, 4) - batch size 2, 4 features
inputs = torch.randn(2, 4, dtype=torch.float32)
targets = torch.randint(0, 2, (2, 4), dtype=torch.float32)

ARGS = {
    'inputs': inputs,
    'targets': targets,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 2) — shape (2, 2) 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
calculate_uncertainty Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
slonetime/EBSeg/ebseg/model/criterion.py
code served (permissive licence) · get_code("2dcb8123d89bb1ff")
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 4, dtype family float, shape (2, 1, 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 torch

torch.manual_seed(0)

# logits should be a tensor of shape (R, 1, ...) where R is number of predictions
# The function expects logits.shape[1] == 1
# Let's create a tensor with shape (2, 1, 4, 3) - R=2, C=1, and spatial dims 4x3
logits = torch.randn(2, 1, 4, 3)

ARGS = {
    "logits": logits
}
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 4, dtype family float, shape (2, 1, 4, 3) — shape (2, 1, 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
dice_loss Ran
ran; our contract draft was wrong, not the code
slonetime/EBSeg/ebseg/model/criterion.py
code served (permissive licence) · get_code("89f75e54ff128be0")
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 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)

# Create small tensors for inputs and targets
# inputs should be arbitrary shape that can be flattened
inputs = torch.randn(2, 4, requires_grad=False)  # batch_size=2, features=4
targets = torch.randint(0, 2, (2, 4), dtype=torch.float32)  # same shape as inputs, binary labels

# num_masks is a float representing the number of masks for normalization
num_masks = 2.0

ARGS = {
    "inputs": inputs,
    "targets": targets,
    "num_masks": num_masks,
}
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 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_predefined_templates Ran
ran; our contract draft was wrong, not the code
slonetime/EBSeg/ebseg/model/clip_utils/utils.py
code served (permissive licence) · get_code("c266c766b74bafca")
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 torch
import numpy as np
from typing import List

torch.manual_seed(0)

# The function takes a single parameter: template_set_name
# It checks if the name exists in PREDEFINED_TEMPLATES dict
# and returns the corresponding list of template strings

ARGS = {
    'template_set_name': 'imagenet'
}
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 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
mask_pooling Ran
ran; our contract draft was wrong, not the code
slonetime/EBSeg/ebseg/model/additional_modules/mask_pooling.py
code served (permissive licence) · get_code("f1a802d7256b1438")
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, 3, 6) 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)

# mask: shape (batch, num_masks, height, width) - boolean or float tensor
# embedding: shape (batch, channels, height, width) - float tensor

batch_size = 2
num_masks = 3
height = 4
width = 5
channels = 6

# Create mask tensor - float values that will be passed through sigmoid and thresholded
mask = torch.randn(batch_size, num_masks, height, width)

# Create embedding tensor
embedding = torch.randn(batch_size, channels, height, width)

ARGS = {
    "mask": mask,
    "embedding": embedding,
}
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, 3, 6) — shape (2, 3, 6) 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
resize_pos_embed2d Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
slonetime/EBSeg/ebseg/model/clip_utils/clip_surgery_visual.py
code served (permissive licence) · get_code("a03c560e737da2c5")
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 (1, 49, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Create a positional embedding tensor [N, L, C]
# N=1 (batch), L=num_prefix_tokens + grid_tokens, C=4 (embedding dim)
# src_shape is (4, 3) so grid has 4*3=12 tokens
# With num_prefix_tokens=1, total L = 1 + 12 = 13
posemb = torch.randn(1, 13, 4)

# Source shape (height, width) for the spatial grid
src_shape = (4, 3)

# Target shape (height, width) for interpolation
tgt_shape = (8, 6)

# Number of prefix tokens (like [CLS] token)
num_prefix_tokens = 1

# Interpolation mode
interpolation = 'bicubic'

# Whether to use antialiasing
antialias = False

ARGS = {
    'posemb': posemb,
    'src_shape': src_shape,
    'tgt_shape': tgt_shape,
    'num_prefix_tokens': num_prefix_tokens,
    'interpolation': interpolation,
    'antialias': antialias,
}
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 (1, 49, 4) — shape (1, 49, 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
sigmoid_ce_loss Ran
ran; our contract draft was wrong, not the code
slonetime/EBSeg/ebseg/model/criterion.py
code served (permissive licence) · get_code("d0c61e8dba511aa3")
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 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
import torch.nn.functional as F

torch.manual_seed(0)

# Create small tensors for inputs and targets
# inputs: predictions (logits), arbitrary shape
# targets: binary labels (0 or 1), same shape as inputs
batch_size = 2
num_elements = 4

inputs = torch.randn(batch_size, num_elements, dtype=torch.float32)
targets = torch.randint(0, 2, (batch_size, num_elements), dtype=torch.float32)

# num_masks: used as divisor in the loss computation
num_masks = 2.0

ARGS = {
    "inputs": inputs,
    "targets": targets,
    "num_masks": num_masks,
}
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 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
batch_sigmoid_ce_loss Not yet run
unverified (harvested, no recorded run)
slonetime/EBSeg/ebseg/model/matcher.py
code served (permissive licence) · get_code("1edd24985036b0bf")
How it ran

no invocation recorded.

build_sam_vit_b Not yet run
unverified (harvested, no recorded run)
slonetime/EBSeg/ebseg/model/segment_anything/build_sam.py
code served (permissive licence) · get_code("91c965b9c51a37f7")
How it ran

no invocation recorded.

build_sam_vit_h Not yet run
unverified (harvested, no recorded run)
slonetime/EBSeg/ebseg/model/segment_anything/build_sam.py
code served (permissive licence) · get_code("8590c79b1c4f7609")
How it ran

no invocation recorded.

build_sam_vit_l Not yet run
unverified (harvested, no recorded run)
slonetime/EBSeg/ebseg/model/segment_anything/build_sam.py
code served (permissive licence) · get_code("dfc513fa0096086b")
How it ran

no invocation recorded.

downsample2d Not yet run
unverified (harvested, no recorded run)
slonetime/EBSeg/ebseg/model/clip_utils/clip_surgery_visual.py
code served (permissive licence) · get_code("5d474b8fcf4cfbb2")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 6 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 0 the same month, 6 later) and 6 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

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

FunctionAlso present inEarliest-datedOthers named
dice_loss 46 other papers
37 dated earlier, 8 later, 1 same year, order not determinable
2112.01527 (2021-12)
Masked-attention Mask Transformer for Universal Image Segmentation
2112.10764 (2021-12)
Mask2Former for Video Instance Segmentation
2203.03605 (2022-03)
DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection
and 43 more not named here
sigmoid_ce_loss 45 other papers
37 dated earlier, 7 later, 1 same year, order not determinable
2112.01527 (2021-12)
Masked-attention Mask Transformer for Universal Image Segmentation
2112.10764 (2021-12)
Mask2Former for Video Instance Segmentation
2203.03605 (2022-03)
DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection
and 42 more not named here
batch_dice_loss 44 other papers
33 dated earlier, 10 later, 1 same year, order not determinable
2112.01527 (2021-12)
Masked-attention Mask Transformer for Universal Image Segmentation
2112.10764 (2021-12)
Mask2Former for Video Instance Segmentation
2203.03605 (2022-03)
DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection
and 41 more not named here
batch_sigmoid_ce_loss 44 other papers
33 dated earlier, 10 later, 1 same year, order not determinable
2112.01527 (2021-12)
Masked-attention Mask Transformer for Universal Image Segmentation
2112.10764 (2021-12)
Mask2Former for Video Instance Segmentation
2203.03605 (2022-03)
DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection
and 41 more not named here
calculate_uncertainty 34 other papers
28 dated earlier, 4 later, 2 same year, order not determinable
2112.01527 (2021-12)
Masked-attention Mask Transformer for Universal Image Segmentation
2112.10764 (2021-12)
Mask2Former for Video Instance Segmentation
2203.04187 (2022-03)
RankSeg: Adaptive Pixel Classification with Image Category Ranking for Segmentation
and 31 more not named here
get_predefined_templates 2 other papers
1 dated earlier, 1 later
2302.12242 (2023-02)
Side Adapter Network for Open-Vocabulary Semantic Segmentation
2407.08268 (2024-07)
Explore the Potential of CLIP for Training-Free Open Vocabulary Semantic Segmentation

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-vocabulary semantic segmentation is a challenging task, which requires the model to output semantic masks of an image beyond a close-set vocabulary. Although many efforts have been made to utilize powerful CLIP models to accomplish this task, they are still easily overfitting to training classes due to the natural gaps in semantic information between training and new classes. To overcome this challenge, we propose a novel framework for openvocabulary semantic segmentation called EBSeg, incorporating an Adaptively Balanced Decoder (AdaB Decoder) and a Semantic Structure Consistency loss (SSC Loss). The AdaB Decoder is designed to generate different image embeddings for both training and new classes. Subsequently, these two types of embeddings are adaptively balanced to fully exploit their ability to recognize training classes and generalization ability for new classes. To learn a consistent semantic structure from CLIP, the SSC Loss aligns the inter-classes affinity in the image feature space with that in the text feature space of CLIP, thereby improving the generalization ability of our model. Furthermore, we employ a frozen SAM image encoder to complement the spatial information that CLIP features lack due to the low training image resolution and image-level supervision inherent in CLIP. Extensive experiments conducted across various benchmarks demonstrate that the proposed EBSeg outperforms the state-of-the-art methods. Our code and trained models will be here: https://github.com/slonetime/EBSeg.

For agents

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

get_harvested_code_for_paper("2406.09829")
get_code_for_paper("2406.09829")
have("2406.09829")

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

curl https://syntology.ai/api/ran/2406.09829.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.09829.svg)](https://syntology.ai/paper/2406.09829)

Connect an agent — have() is free.