Pith. sign in

REVIEW 5 major objections 5 minor 73 references

A Survey on Training-free Open-Vocabulary Semantic Segmentation

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This survey claims to be the first exhaustive overview of training-free open-vocabulary semantic segmentation, organizing over 30 methods into three branches and comparing them on eight benchmarks.

desk verdict Useful taxonomy and writeup of training-free OVSS, but the quantitative rankings and the 'larger backbones hurt' claim outrun the uncontrolled mIoU data. read the letter →

arxiv 2505.22209 v1 pith:ESHBGX3S submitted 2025-05-28 cs.CV

Naomi Kombol , Ivan Martinović , Siniša Šegvić This is my paper
classification cs.CV
keywords semanticsegmentationopen-vocabularytraining-freeCLIPvision-languagemodelsvisualfoundationgenerativesurvey
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that training-free open-vocabulary semantic segmentation—assigning pixel labels to arbitrary classes without any fine-tuning—is now a coherent research area with a clear intellectual structure. It organizes more than 30 methods into three branches: purely CLIP-based approaches that rewire the model's attention, approaches that borrow spatial knowledge from auxiliary visual foundation models, and approaches that use generative models to produce prototypes or attention maps. The survey's central assertion is that this list is exhaustive and its comparison table is trustworthy enough to reveal which design choices matter. A reader would care because the field's promises—arbitrary categories, no training cost, instant deployment—hinge on whether the map and the numbers hold up.

What carries the argument

The organizing device is a three-branch taxonomy, built around how each method obtains dense predictions from a frozen CLIP model. Pure CLIP methods modify inter-token attention (for example by computing similarity in the same semantic space instead of the standard query-key product) or use intermediate layers and algorithmically obtained masks. A second branch adds visual foundation models such as DINO or SAM to refine attention or generate masks for pooling CLIP features. The third branch uses generative models, typically Stable Diffusion, to create class prototypes or cross-attention maps that guide segmentation. The survey also maintains a comparative mIoU table across eight benchmarks as the quantitative backbone of its claims.

What would settle it

Re-running every method listed in the survey on the same benchmark suite with an identical protocol (same backbone, prompts, and metric) and finding that the relative ordering or the larger-backbone degradation reported in Tables 1 and 2 does not reproduce.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that the entire training-free open-vocabulary segmentation literature can be organized as variations on one recipe: take a frozen contrastive vision-language model such as CLIP, alter its last-layer attention or pooling to produce dense per-patch embeddings, and optionally inject spatial structure from a visual foundation model or class prototypes from a generative model. The survey claims this recipe produces full-image segmentation with no additional training cost, and that the performance leaders at present are methods that pool CLIP features over masks produced by SAM or DINO. It also reports that some purely CLIP-based methods get worse with larger backbones, which it reads as evidence about how CLIP's embedding space is used in large models.

Load-bearing premise

The comparison and the field-level observations assume that the mIoU numbers quoted from different original papers were measured under compatible evaluation conditions—same prompts, preprocessing, and post-processing—so that the rankings reflect the methods rather than the measurement pipelines.

Editorial extensions

If this is right

  • Training-free methods can segment arbitrary visual categories at inference time with no additional training, making them deployable where annotated data is scarce.
  • The three-branch taxonomy predicts where future improvements are likely to come from: better attention rewiring, better auxiliary masks, or better generative prototypes.
  • According to the survey, mask-pooling over VFM-generated masks currently yields the strongest results, suggesting spatial priors matter more than further CLIP surgery.
  • The reported degradation of some CLIP-only methods with larger backbones implies that scaling the visual encoder does not automatically improve dense prediction under these recipes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the mIoU table were re-measured under a single evaluation protocol, the relative ranking might shift; the survey itself notes that numbers come from different source papers, so the ranking should be treated as approximate until unified benchmarks exist.
  • The 'larger backbone hurts' observation raises a testable hypothesis: CLIP's token semantics in larger ViTs are more globally concentrated, so attention-rewiring methods lose local discriminability; this could be probed by analyzing token-wise entropy per layer.
  • The taxonomy suggests an untried combination: using a convolutional CLIP backbone with VFM-generated masks, since no existing method explores convolutional CLIP for this task.
  • Generative methods that create class prototypes offline pay a setup cost that is not counted as training; a fair cost comparison should include prototype generation time when claiming 'training-free'.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. This manuscript presents a survey of training-free open-vocabulary semantic segmentation methods, organizing them into three branches: purely CLIP-based approaches, approaches using auxiliary visual foundation models, and approaches using generative models. It includes preliminaries on vision-language models and visual foundation models, detailed descriptions of over 30 methods, quantitative mIoU comparisons on eight benchmarks, and a discussion of limitations and future directions.

Significance. If the quantitative comparisons were reliable, the survey would be a useful entry point and reference for a rapidly growing area. The taxonomy is sensible, the method descriptions are generally accessible, and the historical organization helps trace the development of ideas. The paper also identifies several underexplored directions, such as ConvNeXt CLIP and single-pass inference. However, the load-bearing quantitative claims are not adequately supported by the data as presented, and several methods lack proper citations, so the survey requires revision before it can serve as a trustworthy reference.

major comments (5)
  1. [Section 4.1, Table 1, Section 4.4, Table 2] The quantitative comparisons in Tables 1 and 2 are not sufficiently controlled to support the rankings and backbone-size conclusions. Table 1 mixes results from original papers with results reproduced by SCLIP and ITACLIP, and uses different backbones across rows (ViT-L/14 for CaR, ReCO, and TAG; ViT-H/14 for Trident). Consequently, the claim in Section 4.1 that DBA-CLIP and CorrCLIP are 'current leaders' and the claim in Section 4.4 that ClearCLIP, GEM, ITACLIP, ResCLIP, and CASS 'suffer degraded performance when using larger backbone models' may be artifacts of differing evaluation protocols rather than true properties of the methods. Please either run or cite a controlled comparison under a single protocol, or substantially qualify these conclusions and remove explicit rankings.
  2. [Section 3.1.1] The method 'TagCLIP' is described in detail but no citation is given in the text. Table 1 cites TagCLIP to reference [2] (ITACLIP), which is described as a reproduction, not the original source. The original TagCLIP paper should be cited, and the relationship between TagCLIP and ITACLIP's reproduction should be clarified.
  3. [Section 3.3] CLIPer is described without a citation, even though reference [56] (Sun et al., 2024) appears to be the CLIPer paper. Add the citation at the first mention of CLIPer in Section 3.3.
  4. [References [66] and [67], Sections 3.2.2 and 4.1] References [66] and [67] are the same paper (Yang and Gong, 'Tuning-free universally-supervised semantic segmentation', 2024), but they are listed as separate entries and cited differently in Sections 3.2.2 and 4.1. Merge the duplicate entry and standardize the citation to avoid confusion.
  5. [Table 2] Table 2 is introduced as a comparison 'with respect to backbone size,' but rows labeled 'NACLIP - PAMR' and 'ITACLIP - PAMR' appear to be post-processing variants rather than backbone changes. The footnotes do not specify the backbone for these rows, so it is unclear what the comparison in Section 4.4 actually shows. Please remove these rows or state explicitly that they are not backbone comparisons.
minor comments (5)
  1. [Section 2.2] The text uses both 'Visual Foundation Models (VMF-s)' and 'VFM-s' in the same paragraph; please standardize the abbreviation.
  2. [Section 3.3 vs. Table 1] The method is spelled 'DiffSegmentor' in Section 3.3 but 'DiffSegmenter' in Table 1; use consistent spelling throughout.
  3. [References [46] and [47]] References [46] and [47] are duplicate entries, both listing Rombach et al., 'High-resolution image synthesis with latent diffusion models.' Remove one.
  4. [Section 4.2] Section 4.2 shifts from describing how different methods handle the background class to proposing a research direction without a clear transition, making the paragraph difficult to follow; consider restructuring it.
  5. [Table 1] The row label 'CaRL/14*' is confusing; clarify whether the asterisk refers to the footnote about ViT-L/14 and whether the method also uses a ViT-B/16 instance.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the survey compiles external published results without fitting parameters, making predictions, or relying on self-citations.

full rationale

This paper is a survey and makes no derivations, fits no parameters, and presents no first-principles result. Its central claims are taxonomic and descriptive: it organizes over 30 training-free open-vocabulary segmentation methods into categories and discusses their ideas. The quantitative comparison in Tables 1 and 2 imports mIoU numbers from the original papers or from prior reproductions, as explicitly stated in the Table 1 footnote ('Unless indicated with a superscript, the results were imported from the original papers. The baseline CLIP, MaskCLIP, and ReCO evaluations were instead taken from [60]... TagCLIP results are from the reproduction in [2]'). Importing numbers and then ranking methods is not circular: the survey does not use those numbers to define the methods, and the ranking is simply a restatement of the imported measurements rather than a prediction derived from them. The Section 4.4 observation that some methods 'suffer degraded performance when using larger backbone models' is an empirical generalization over imported numbers and, while potentially confounded by heterogeneous evaluation protocols, that is a data-integrity concern, not a circularity concern. No author self-citations appear; the authors cite external work throughout. Therefore no load-bearing step reduces to its own inputs, and the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

This is a survey, so there are no fitted parameters or invented entities. The load-bearing assumptions are the exhaustiveness of the taxonomy and the comparability and accuracy of the reported benchmark numbers; both are assumptions about external literature rather than quantities derived in this paper.

assumptions (3)
  • domain assumption The three-branch taxonomy (purely CLIP-based, CLIP with visual foundation models, CLIP with generative methods) is exhaustive for training-free OVSS.
    Section 3 defines the structure of the survey and Section 1 calls it exhaustive. If a significant branch or family of methods exists outside these categories, the survey's central coverage claim fails. The paper provides no systematic search protocol to justify exhaustiveness.
  • domain assumption The mIoU values in Tables 1 and 2 are accurate and comparable as reported.
    Table 1's footnote states that most results were imported from original papers and some baselines were taken from [60]. The comparisons in Sections 4.1 and 4.4 assume these numbers are faithful and measured under comparable settings.
  • domain assumption The cited external models (CLIP, DINO, DINOv2, SAM, Stable Diffusion) behave as described in their source papers.
    Section 2 summarizes these models, and the method descriptions rely on the reader accepting those summaries as accurate enough for the survey's conclusions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Survey on Training-free Open-Vocabulary Semantic Segmentation." pith.science (2026). https://pith.science/paper/ESHBGX3S

@misc{pith2026250522209,
  author       = {Pith},
  title        = {Pith review of: A Survey on Training-free Open-Vocabulary Semantic Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ESHBGX3S}},
  note         = {Machine review of arXiv:2505.22209}
}
read the original abstract

Semantic segmentation is one of the most fundamental tasks in image understanding with a long history of research, and subsequently a myriad of different approaches. Traditional methods strive to train models up from scratch, requiring vast amounts of computational resources and training data. In the advent of moving to open-vocabulary semantic segmentation, which asks models to classify beyond learned categories, large quantities of finely annotated data would be prohibitively expensive. Researchers have instead turned to training-free methods where they leverage existing models made for tasks where data is more easily acquired. Specifically, this survey will cover the history, nuance, idea development and the state-of-the-art in training-free open-vocabulary semantic segmentation that leverages existing multi-modal classification models. We will first give a preliminary on the task definition followed by an overview of popular model archetypes and then spotlight over 30 approaches split into broader research branches: purely CLIP-based, those leveraging auxiliary visual foundation models and ones relying on generative methods. Subsequently, we will discuss the limitations and potential problems of current research, as well as provide some underexplored ideas for future study. We believe this survey will serve as a good onboarding read to new researchers and spark increased interest in the area.

Figures

Figures reproduced from arXiv: 2505.22209 by the authors.

Figure 1
Figure 1. Overview of Training-free Open-Vocabulary Semantic [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Modified ViT encoder block with removed MLP and [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. Standard ViT encoder block that first normalizes inputs [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison on the Pascal VOC21 [ [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

73 extracted references · 52 canonical work pages

  1. [2]

    Arda Aydın, Efe Mert C ¸ ırpar, Elvin Abdinli, Gozde Unal, and Yusuf H

    M. Arda Aydın, Efe Mert C ¸ ırpar, Elvin Abdinli, Gozde Unal, and Yusuf H. Sahin. Itaclip: Boosting training-free seman- tic segmentation with image, text, and architectural enhance- ments, 2024. 3, 4, 7, 8

    work page 2024

  2. [56]

    Cliper: Hierarchically improving spatial represen- tation of clip for open-vocabulary semantic segmentation,

    Lin Sun, Jiale Cao, Jin Xie, Xiaoheng Jiang, and Yanwei Pang. Cliper: Hierarchically improving spatial represen- tation of clip for open-vocabulary semantic segmentation,

    work page

  3. [66]

    Tuning-free universally- supervised semantic segmentation, 2024

    Xiaobo Yang and Xiaojin Gong. Tuning-free universally- supervised semantic segmentation, 2024. 4, 5

    work page 2024

  4. [67]

    Tuning-free universally- supervised semantic segmentation, 2024

    Xiaobo Yang and Xiaojin Gong. Tuning-free universally- supervised semantic segmentation, 2024. 5, 6

    work page 2024

  5. [1]

    Single-Stage Semantic Segmentation from Image Labels

    Nikita Araslanov and Stefan Roth. Single-stage semantic segmentation from image labels. CoRR, abs/2005.08104,

    work page Pith review arXiv 2005

  6. [3]

    Self-calibrated clip for training-free open-vocabulary segmentation, 2024

    Sule Bai, Yong Liu, Yifei Han, Haoji Zhang, and Yansong Tang. Self-calibrated clip for training-free open-vocabulary segmentation, 2024. 3, 4, 6

    work page 2024

  7. [4]

    Fossil: Free open-vocabulary semantic seg- mentation through synthetic references retrieval

    Luca Barsellotti, Roberto Amoroso, Lorenzo Baraldi, and Rita Cucchiara. Fossil: Free open-vocabulary semantic seg- mentation through synthetic references retrieval. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 1464–1473, 2024. 5, 6

    work page 2024

  8. [5]

    Training-free open- vocabulary segmentation with offline diffusion-augmented prototype generation, 2024

    Luca Barsellotti, Roberto Amoroso, Marcella Cornia, Lorenzo Baraldi, and Rita Cucchiara. Training-free open- vocabulary segmentation with offline diffusion-augmented prototype generation, 2024. 5, 6

    work page 2024

Show all 73 references
  1. [6]

    Grounding everything: Emerging localiza- tion properties in vision-language transformers, 2023

    Walid Bousselham, Felix Petersen, Vittorio Ferrari, and Hilde Kuehne. Grounding everything: Emerging localiza- tion properties in vision-language transformers, 2023. 3, 5, 8

    2023

  2. [7]

    Breunig, Hans-Peter Kriegel, Raymond T

    Markus M. Breunig, Hans-Peter Kriegel, Raymond T. Ng, and J ¨org Sander. Lof: identifying density-based local out- liers. In Proceedings of the 2000 ACM SIGMOD Interna- tional Conference on Management of Data , page 93–104, New York, NY , USA, 2000. Association for Computing M...

    2000

  3. [8]

    Zero-shot semantic segmentation

    Maxime Bucher, Tuan-Hung Vu, Matthieu Cord, and Patrick P ´erez. Zero-shot semantic segmentation. CoRR, abs/1906.00817, 2019. 1, 2

    1906 arXiv

  4. [9]

    Holger Caesar, Jasper R. R. Uijlings, and Vittorio Ferrari. Coco-stuff: Thing and stuff classes in context. CoRR, abs/1612.03716, 2016. 7, 8

    2016 arXiv

  5. [10]

    Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom

    Holger Caesar, Varun Bankiti, Alex H. Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. nuscenes: A multimodal dataset for autonomous driving. CoRR, abs/1903.11027, 2019. 1

    1903 arXiv

  6. [11]

    A computational approach to edge detection

    John Canny. A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, PAMI-8(6):679–698, 1986. 1

    1986

  7. [12]

    Emerg- ing properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herv ´e J´egou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers. CoRR, abs/2104.14294, 2021. 2, 5, 6

    2021 arXiv

  8. [13]

    Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L. Yuille. Deeplab: Semantic im- age segmentation with deep convolutional nets, atrous con- volution, and fully connected crfs. IEEE Transactions on Pattern Analysis and Machine Intelligence , 40(4):834–848,

  9. [14]

    Schwing, Alexan- der Kirillov, and Rohit Girdhar

    Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexan- der Kirillov, and Rohit Girdhar. Masked-attention mask transformer for universal image segmentation. CoRR, abs/2112.01527, 2021. 1

    2021 arXiv

  10. [15]

    Schwing, and Alexander Kir- illov

    Bowen Cheng, Alexander G. Schwing, and Alexander Kir- illov. Per-pixel classification is not all you need for semantic segmentation. CoRR, abs/2107.06278, 2021. 1

    2021 arXiv

  11. [16]

    The cityscapes dataset for semantic urban scene understanding

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. CoRR, abs/1604.01685, 2016. 7, 8

    2016 arXiv

  12. [17]

    Freeseg-diff: Training-free open-vocabulary segmentation with diffusion models, 2024

    Barbara Toniella Corradini, Mustafa Shukor, Paul Couairon, Guillaume Couairon, Franco Scarselli, and Matthieu Cord. Freeseg-diff: Training-free open-vocabulary segmentation with diffusion models, 2024. 5, 6

    2024

  13. [18]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2010 arXiv

  14. [19]

    A density-based algorithm for discovering clusters in large spatial databases with noise

    Martin Ester, Hans-Peter Kriegel, J ¨org Sander, and Xiaowei Xu. A density-based algorithm for discovering clusters in large spatial databases with noise. In Proceedings of the Sec- ond International Conference on Knowledge Discovery and Data Mining, page 226–231. AAAI Press, 1996. 4

    1996

  15. [20]

    The pascal visual object classes (voc) challenge

    Mark Everingham, Luc Van Gool, Christopher Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge. International Journal of Computer Vision, 88:303–338, 2010. 5, 7, 8

    2010

  16. [21]

    Pay attention to your neighbours: Training-free open-vocabulary semantic segmentation, 2024

    Sina Hajimiri, Ismail Ben Ayed, and Jose Dolz. Pay attention to your neighbours: Training-free open-vocabulary semantic segmentation, 2024. 3

    2024

  17. [22]

    Girshick

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross B. Girshick. Mask R-CNN. CoRR, abs/1703.06870, 2017. 1

    2017 arXiv

  18. [23]

    Girshick

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross B. Girshick. Masked autoencoders are scal- able vision learners. CoRR, abs/2111.06377, 2021. 5

    2021 arXiv

  19. [24]

    Zemel, and M.A

    Xuming He, R.S. Zemel, and M.A. Carreira-Perpinan. Mul- tiscale conditional random fields for image labeling. In Pro- ceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004., pages II–II, 2004. 1

    2004

  20. [25]

    Open-vocabulary instance segmentation via ro- bust cross-modal pseudo-labeling

    Dat Huynh, Jason Kuen, Zhe Lin, Jiuxiang Gu, and Ehsan Elhamifar. Open-vocabulary instance segmentation via ro- bust cross-modal pseudo-labeling. CoRR, abs/2111.12698,

    arXiv

  21. [26]

    Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V . Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig. Scaling up visual and vision-language representation learning with noisy text supervision.CoRR, abs/2102.05918,

    arXiv

  22. [27]

    In defense of lazy vi- sual grounding for open-vocabulary semantic segmentation,

    Dahyun Kang and Minsu Cho. In defense of lazy vi- sual grounding for open-vocabulary semantic segmentation,

  23. [28]

    Diffusion models for open-vocabulary segmen- tation, 2024

    Laurynas Karazija, Iro Laina, Andrea Vedaldi, and Christian Rupprecht. Diffusion models for open-vocabulary segmen- tation, 2024. 5 9

    2024

  24. [29]

    Maskdiffusion: Exploiting pre-trained diffusion models for semantic seg- mentation, 2024

    Yasufumi Kawano and Yoshimitsu Aoki. Maskdiffusion: Exploiting pre-trained diffusion models for semantic seg- mentation, 2024. 5, 6

    2024

  25. [30]

    Tag: Guidance-free open-vocabulary semantic segmentation, 2024

    Yasufumi Kawano and Yoshimitsu Aoki. Tag: Guidance-free open-vocabulary semantic segmentation, 2024. 4, 5

    2024

  26. [31]

    Distilling spectral graph for object-context aware open-vocabulary semantic segmenta- tion, 2024

    Chanyoung Kim, Dayun Ju, Woojung Han, Ming-Hsuan Yang, and Seong Jae Hwang. Distilling spectral graph for object-context aware open-vocabulary semantic segmenta- tion, 2024. 4, 5, 8

    2024

  27. [32]

    Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross Girshick

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C. Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross Girshick. Segment anything, 2023. 2, 5, 6

    2023

  28. [33]

    Clearclip: Decom- posing clip representations for dense vision-language infer- ence, 2024

    Mengcheng Lan, Chaofeng Chen, Yiping Ke, Xinjiang Wang, Litong Feng, and Wayne Zhang. Clearclip: Decom- posing clip representations for dense vision-language infer- ence, 2024. 3, 8

    2024

  29. [34]

    Proxyclip: Proxy attention improves clip for open-vocabulary segmentation,

    Mengcheng Lan, Chaofeng Chen, Yiping Ke, Xinjiang Wang, Litong Feng, and Wayne Zhang. Proxyclip: Proxy attention improves clip for open-vocabulary segmentation,

  30. [35]

    A closer look at the explainability of contrastive language-image pre-training, 2024

    Yi Li, Hualiang Wang, Yiqun Duan, Jiheng Zhang, and Xi- aomeng Li. A closer look at the explainability of contrastive language-image pre-training, 2024. 1, 3

    2024

  31. [36]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feicht- enhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. CoRR, abs/2201.03545, 2022. 8

    2022 arXiv

  32. [37]

    Fully convolutional networks for semantic segmentation

    Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. CoRR, abs/1411.4038, 2014. 1

    2014 arXiv

  33. [38]

    Segment anything in medical images

    Jun Ma, Yuting He, Feifei Li, Lin Han, Chenyu You, and Bo Wang. Segment anything in medical images. Nature Communications, 15(1), 2024. 1

    2024

  34. [39]

    Meyer and S

    F. Meyer and S. Beucher. Morphological segmentation. Journal of Visual Communication and Image Representation, 1(1):21–46, 1990. 1

    1990

  35. [40]

    The role of context for object detection and se- mantic segmentation in the wild

    Roozbeh Mottaghi, Xianjie Chen, Xiaobai Liu, Nam-Gyu Cho, Seong-Whan Lee, Sanja Fidler, Raquel Urtasun, and Alan Yuille. The role of context for object detection and se- mantic segmentation in the wild. In 2014 IEEE Conference on Computer Vision and Pattern Recognition , pages...

    2014

  36. [41]

    Emerdiff: Emerging pixel-level semantic knowledge in diffusion models, 2024

    Koichi Namekata, Amirmojtaba Sabour, Sanja Fidler, and Seung Wook Kim. Emerdiff: Emerging pixel-level semantic knowledge in diffusion models, 2024. 5

    2024

  37. [42]

    Dinov2: Learning robust visual features with- out supervision, 2024

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mah- moud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michae...

    2024

  38. [43]

    A threshold selection method from gray- level histograms

    Nobuyuki Otsu. A threshold selection method from gray- level histograms. IEEE Transactions on Systems, Man, and Cybernetics, 9(1):62–66, 1979. 1

    1979

  39. [44]

    Cattle segmentation and contour extraction based on mask r-cnn for precision livestock farming

    Yongliang Qiao, Matthew Truman, and Salah Sukkarieh. Cattle segmentation and contour extraction based on mask r-cnn for precision livestock farming. Computers and Elec- tronics in Agriculture, 165:104958, 2019. 1

    2019

  40. [45]

    Learning transferable vi- sual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable vi- sual models from natural language supervision. CoRR, abs/2103.000...

    2021 arXiv

  41. [47]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. CoRR, abs/2112.10752,

    arXiv

  42. [48]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. CoRR, abs/1505.04597, 2015. 1

    2015 arXiv

  43. [49]

    Deep learning based real-time industrial framework for rotten and fresh fruit detection using semantic segmentation

    Kyamelia Roy, Sheli Chaudhuri, and Sayan Pramanik. Deep learning based real-time industrial framework for rotten and fresh fruit detection using semantic segmentation. Microsys- tem Technologies, 27:1–11, 2021. 1

    2021

  44. [50]

    Selvaraju, Abhishek Das, Ramakrishna Vedantam, Michael Cogswell, Devi Parikh, and Dhruv Ba- tra

    Ramprasaath R. Selvaraju, Abhishek Das, Ramakrishna Vedantam, Michael Cogswell, Devi Parikh, and Dhruv Ba- tra. Grad-cam: Why did you say that? visual explanations from deep networks via gradient-based localization. CoRR, abs/1610.02391, 2016. 4

    arXiv

  45. [51]

    Ex- plore the potential of clip for training-free open vocabulary semantic segmentation, 2024

    Tong Shao, Zhuotao Tian, Hang Zhao, and Jingyong Su. Ex- plore the potential of clip for training-free open vocabulary semantic segmentation, 2024. 3, 4

    2024

  46. [52]

    Jianbo Shi and J. Malik. Normalized cuts and image segmen- tation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(8):888–905, 2000. 1

    2000

  47. [53]

    Harnessing vi- sion foundation models for high-performance, training-free open vocabulary segmentation, 2024

    Yuheng Shi, Minjing Dong, and Chang Xu. Harnessing vi- sion foundation models for high-performance, training-free open vocabulary segmentation, 2024. 4, 5

    2024

  48. [54]

    Reco: Re- trieve and co-segment for zero-shot transfer, 2022

    Gyungin Shin, Weidi Xie, and Samuel Albanie. Reco: Re- trieve and co-segment for zero-shot transfer, 2022. 4

    2022

  49. [55]

    Unsupervised object local- ization: Observing the background to discover objects, 2023

    Oriane Sim ´eoni, Chlo´e Sekkat, Gilles Puy, Antonin V obecky, ´Eloi Zablocki, and Patrick P´erez. Unsupervised object local- ization: Observing the background to discover objects, 2023. 4

    2023

  50. [57]

    Clip as rnn: Segment countless visual concepts without training endeavor, 2024

    Shuyang Sun, Runjia Li, Philip Torr, Xiuye Gu, and Siyang Li. Clip as rnn: Segment countless visual concepts without training endeavor, 2024. 3, 4, 6

    2024

  51. [58]

    Lema, Oscar D

    Rub ´en Usamentiaga, Dar´ıo G. Lema, Oscar D. Pedrayes, and Daniel F. Garcia. Automated surface defect detection in met- als: A comparative review of object detection and semantic segmentation using deep learning. IEEE Transactions on In- dustry Applications, 58(3):4203–4213, ...

    2022

  52. [59]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. CoRR, abs/1706.03762, 2017. 2

    2017 arXiv

  53. [60]

    Sclip: Rethinking self-attention for dense vision-language inference, 2024

    Feng Wang, Jieru Mei, and Alan Yuille. Sclip: Rethinking self-attention for dense vision-language inference, 2024. 1, 3, 4, 5, 7

    2024

  54. [61]

    Diffusion model is secretly a training-free open vocabulary semantic segmenter,

    Jinglong Wang, Xiawei Li, Jing Zhang, Qingyuan Xu, Qin Zhou, Qian Yu, Lu Sheng, and Dong Xu. Diffusion model is secretly a training-free open vocabulary semantic segmenter,

  55. [62]

    Tokencut: Segmenting objects in images and videos with self-supervised transformer and normalized cut, 2023

    Yangtao Wang, Xi Shen, Yuan Yuan, Yuming Du, Maomao Li, Shell Xu Hu, James L Crowley, and Dominique Vaufrey- daz. Tokencut: Segmenting objects in images and videos with self-supervised transformer and normalized cut, 2023. 2

    2023

  56. [63]

    Image-to-image matching via foundation models: A new perspective for open-vocabulary semantic segmentation,

    Yuan Wang, Rui Sun, Naisong Luo, Yuwen Pan, and Tianzhu Zhang. Image-to-image matching via foundation models: A new perspective for open-vocabulary semantic segmentation,

  57. [64]

    Clip-diy: Clip dense infer- ence yields open-vocabulary semantic segmentation for-free,

    Monika Wysocza ´nska, Micha ¨el Ramamonjisoa, Tomasz Trzci´nski, and Oriane Sim ´eoni. Clip-diy: Clip dense infer- ence yields open-vocabulary semantic segmentation for-free,

  58. [65]

    Groupvit: Semantic segmentation emerges from text supervision, 2022

    Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, and Xiaolong Wang. Groupvit: Semantic segmentation emerges from text supervision, 2022. 1, 2

    2022

  59. [68]

    Resclip: Residual attention for training-free dense vision- language inference, 2024

    Yuhang Yang, Jinhong Deng, Wen Li, and Lixin Duan. Resclip: Residual attention for training-free dense vision- language inference, 2024. 3, 4, 8

    2024

  60. [69]

    Advanced agricultural disease image recognition technologies: A re- view

    Yuan Yuan, Lei Chen, Huarui Wu, and Lin Li. Advanced agricultural disease image recognition technologies: A re- view. Information Processing in Agriculture, 9, 2021. 1

    2021

  61. [70]

    Sigmoid loss for language image pre-training,

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training,

  62. [71]

    Corrclip: Recon- structing correlations in clip with off-the-shelf foundation models for open-vocabulary semantic segmentation, 2024

    Dengke Zhang, Fagui Liu, and Quan Tang. Corrclip: Recon- structing correlations in clip with off-the-shelf foundation models for open-vocabulary semantic segmentation, 2024. 4, 5, 6

    2024

  63. [72]

    Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip H. S. Torr, and Li Zhang. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. CoRR, abs/2012.15840, 2020. 1

    2012 arXiv

  64. [73]

    Semantic understand- ing of scenes through the ADE20K dataset

    Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Semantic understand- ing of scenes through the ADE20K dataset. CoRR, abs/1608.05442, 2016. 7, 8

    2016 arXiv

  65. [74]

    Extract free dense labels from clip, 2022

    Chong Zhou, Chen Change Loy, and Bo Dai. Extract free dense labels from clip, 2022. 1, 3, 5, 6 11

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.