Pith. sign in

REVIEW 3 major objections 2 minor 61 references

Correspondence as Video: Test-Time Adaption on SAM2 for Reference Segmentation in the Wild

T0 review · 3 major / 2 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper proposes representing reference-target correspondence as a pseudo video, letting SAM2's interactive video segmentation propagate a reference mask to the target, with reported gains exceeding 5% over state-of-the-art.

desk verdict The abstract describes a clever idea, but the supplied full text is a different paper, so the claimed 5% gain is entirely unverifiable. read the letter →

arxiv 2508.07759 v1 pith:4JSOUZZC submitted 2025-08-11 cs.CV

Haoran Wang , Zekun Li , Jian Zhang , Lei Qi , Yinghuan Shi This is my paper
classification cs.CV
keywords referencesegmentationSAM2pseudovideodiffusion-basedsemantictransitiontest-timegeometricalignmentobjectadaptationCAV-SAM
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 proposes CAV-SAM, which recasts reference segmentation as a video-object-segmentation problem. Instead of meta-learning, it builds a synthetic frame sequence that semantically transitions from the reference image to the target, then runs SAM2's interactive video segmentation to propagate the reference mask across that sequence. Two modules do the work: a diffusion-based module constructs the transition frames, and a test-time alignment module corrects geometric changes. On widely used datasets, the authors report segmentation improvements exceeding 5% over prior methods. The significance would be a lightweight, training-free adaptation route for SAM2 to downstream reference-segmentation tasks.

What carries the argument

The pseudo-video representation built by two modules: Diffusion-Based Semantic Transition (DBST), which uses a diffusion model to generate an ordered semantic transformation sequence from reference to target, and Test-Time Geometric Alignment (TTGA), which fine-tunes alignment at test time to handle geometric changes across the sequence. Together they turn two still images into a video-like input that SAM2's interactive video object segmentation can process.

What would settle it

Run CAV-SAM on a set of reference-target pairs with large viewpoint or appearance changes and measure whether the propagated mask's IoU on the target degrades as the number of diffusion transition steps increases. If the mask drifts off the object at any intermediate step, the identity-preservation premise is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the inherent correspondence between a reference-target image pair can be coded as a pseudo video: a synthetic frame sequence that moves semantically and geometrically from the reference to the target. SAM2, which already performs interactive video object segmentation, can then propagate the reference mask across that sequence to segment the target object. This reframing bypasses meta-learning pipelines, replacing them with a diffusion-based semantic transition module and a test-time geometric alignment module. The reported result is that this pseudo-video formulation improves segmentation performance by more than 5% over state-of-the-art reference-segmentation metho

Load-bearing premise

The diffusion-generated intermediate frames must preserve the identity and correspondence of the reference object well enough that SAM2's mask propagation ends on the target; if those synthetic frames distort or lose the object, the pipeline fails regardless of the alignment module.

Editorial extensions

If this is right

  • Reference segmentation can be performed with SAM2's existing video segmentation capability, avoiding costly meta-training.
  • The method offers a lightweight test-time adaptation route for large vision models on downstream segmentation tasks.
  • The diffusion-generated sequence acts as an explicit semantic bridge, potentially improving robustness to large appearance and pose gaps between reference and target.
  • The reported >5% improvement suggests the pseudo-video formulation is a competitive alternative to meta-learned reference segmentation.
  • Test-time geometric alignment being sufficient indicates that synthetic frame sequences can be corrected on the fly without additional training data.

Reading between the lines

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

  • The pseudo-video idea may extend beyond SAM2 to other interactive segmentors and to other dense correspondence tasks such as matting or depth propagation, wherever reference-target alignment matters.
  • If DBST's intermediate frames preserve object identity well, the method could double as a data-augmentation or uncertainty-probe mechanism: failures in mask propagation would highlight where the semantic transition breaks down.
  • The attached full text describes a different dataset paper and contains no CAV-SAM experiments, ablations, or implementation details; the >5% claim therefore cannot be verified from the provided text alone and rests entirely on the abstract.
  • The dependence on a diffusion model for the transition sequence suggests that the method's ceiling may be tied to diffusion fidelity; cheaper or faster transition generators might be a natural next test.
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

3 major / 2 minor

Summary. The submission, as received, consists of an abstract for a paper titled "Correspondence as Video: Test-Time Adaption on SAM2 for Reference Segmentation in the Wild" and a full-text section that is in fact a different paper, UniSVG (arXiv:2508.07766), on vector-graphics understanding and generation. The abstract describes CAV-SAM, a method that represents reference-target correspondence as a pseudo video, uses a Diffusion-Based Semantic Transition (DBST) module, a Test-Time Geometric Alignment (TTGA) module, and adapts SAM2's interactive video object segmentation to reference segmentation. It claims segmentation improvements exceeding 5% over SOTA methods. None of the CAV-SAM method, its mathematical formulation, implementation details, experimental protocol, datasets, baselines, or results appears anywhere in the submitted text.

Significance. The high-level idea is conceptually attractive: reusing SAM2's interactive video object segmentation through a pseudo-video construction could provide a lightweight alternative to meta-learning for reference segmentation, arguably with broad applicability. If the claimed >5% improvement over SOTA were substantiated, the contribution would be of interest to the computer-vision community. However, in its current form the submission provides no verifiable method or evidence. The novelty, correctness, and experimental validity of CAV-SAM cannot be assessed from the material supplied, so the significance of the claimed result is entirely ungrounded.

major comments (3)
  1. [Full Text] The full text provided is not the CAV-SAM paper described in the abstract; it is the UniSVG paper on vector graphics with different authors, abstract, and contributions. Consequently, the manuscript contains no description of DBST, TTGA, the pseudo-video construction, the adaptation of SAM2, or any related equations or algorithm. This is a load-bearing absence: the central claim cannot be checked against any method description.
  2. [Abstract] The assertion of "segmentation performance improvements exceeding 5% over SOTA methods" is made without specifying datasets, evaluation metrics, baselines, or experimental conditions. No error bars or statistical significance tests are available. As presented in the full text, there is no corresponding experimental section, table, or figure to support this quantitative claim.
  3. [Method (DBST/TTGA)] The core premise that a diffusion-based semantic transition sequence preserves object identity and correspondence well enough for SAM2 to propagate a reference mask across synthetic frames is untested. The abstract gives no details of the diffusion formulation, the construction of the semantic transition, the alignment procedure, or any ablation isolating the contribution of DBST and TTGA. Without this material the correctness of the proposed pipeline cannot be evaluated.
minor comments (2)
  1. [Abstract / Notation] Even within the abstract, the phrase "exceeding 5%" is ambiguous: it could mean a 5-point absolute improvement or a 5% relative improvement, and the baseline(s) "SOTA" are not identified.
  2. [General presentation] The submission lacks a clear statement of contributions, a problem definition, and a section-level organization matching the abstract. The mismatch between the abstract and the full text should be resolved editorially before any review can proceed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity detectable: full text is a different paper (UniSVG), so the CAV-SAM derivation chain is absent and no reduction to inputs can be exhibited.

full rationale

The abstract describes CAV-SAM (Correspondence As Video for SAM) with DBST and TTGA modules and claims a >5% improvement over SOTA, but the supplied full text is arXiv:2508.07766 (UniSVG), a dataset paper on vector graphics understanding and generation. The CAV-SAM method, equations, ablations, and experimental details are entirely absent. Under the stated hard rules, circularity may only be claimed when the paper's own text exhibits a specific reduction (e.g., Eq. X = Eq. Y by construction, or a fitted parameter renamed as a prediction). No such reduction can be quoted for CAV-SAM because the derivation chain is missing. The UniSVG text itself contains no self-referential or circular argument: it constructs a dataset, fine-tunes open-source MLLMs, and reports benchmark results. The mismatch between the abstract and the full text is a serious verification issue (the central claim cannot be checked), but it is not circularity. Therefore the honest finding is score 0 with no circular steps identified.

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

Because only the abstract is available and the full text is from a different paper, no free parameters, hand-chosen constants, or invented entities can be enumerated. The axioms listed are the unstated domain-level premises required for the abstract's method to work.

assumptions (3)
  • domain assumption SAM2's interactive video object segmentation can propagate a mask across a pseudo-video built from a reference-target image pair.
    This is the core enabling premise of CAV-SAM. It is stated implicitly in the abstract's claim that the pseudo-video perspective allows SAM2 to be adapted to reference segmentation.
  • domain assumption A diffusion-based semantic transition sequence preserves enough object correspondence to support accurate mask propagation.
    DBST is introduced in the abstract as the module constructing the semantic transformation sequence. Its faithfulness is assumed, not demonstrated in the abstract.
  • domain assumption Test-time geometric alignment via fine-tuning converges and improves alignment without catastrophic forgetting or overfitting to a single pair.
    TTGA is described as aligning geometric changes through test-time fine-tuning. The abstract does not state its behavior across diverse in-the-wild pairs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Correspondence as Video: Test-Time Adaption on SAM2 for Reference Segmentation in the Wild." pith.science (2026). https://pith.science/paper/4JSOUZZC

@misc{pith2026250807759,
  author       = {Pith},
  title        = {Pith review of: Correspondence as Video: Test-Time Adaption on SAM2 for Reference Segmentation in the Wild},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4JSOUZZC}},
  note         = {Machine review of arXiv:2508.07759}
}
read the original abstract

Large vision models like the Segment Anything Model (SAM) exhibit significant limitations when applied to downstream tasks in the wild. Consequently, reference segmentation, which leverages reference images and their corresponding masks to impart novel knowledge to the model, emerges as a promising new direction for adapting vision models. However, existing reference segmentation approaches predominantly rely on meta-learning, which still necessitates an extensive meta-training process and brings massive data and computational cost. In this study, we propose a novel approach by representing the inherent correspondence between reference-target image pairs as a pseudo video. This perspective allows the latest version of SAM, known as SAM2, which is equipped with interactive video object segmentation (iVOS) capabilities, to be adapted to downstream tasks in a lightweight manner. We term this approach Correspondence As Video for SAM (CAV-SAM). CAV-SAM comprises two key modules: the Diffusion-Based Semantic Transition (DBST) module employs a diffusion model to construct a semantic transformation sequence, while the Test-Time Geometric Alignment (TTGA) module aligns the geometric changes within this sequence through test-time fine-tuning. We evaluated CAVSAM on widely-used datasets, achieving segmentation performance improvements exceeding 5% over SOTA methods. Implementation is provided in the supplementary materials.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 21 canonical work pages

  1. [1]

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Mensch, Katherine Millican, Malcolm Reynolds, et al

  2. [2]

    Anthropic. 2023. Claude 3.7 Sonnet System Card . Technical Report. Anthropic

    work page 2023

  3. [3]

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. 2023. Qwen technical report. arXiv preprint arXiv:2309.16609 (2023)

    arXiv 2023

  4. [4]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, and Junyang Lin. 2025. Qwen2.5-VL Technical Rep...

    arXiv 2025

  5. [5]

    Tom B Brown. 2020. Language models are few-shot learners. arXiv preprint arXiv:2005.14165 (2020)

    arXiv 2020

  6. [6]

    Alexandre Carlier, Martin Danelljan, Alexandre Alahi, and Radu Timofte. 2020. DeepSVG: A Hierarchical Generative Network for Vector Graphics Animation. arXiv:2007.11301 [cs.CV] https://arxiv.org/abs/2007.11301

    arXiv 2020

  7. [7]

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

    arXiv 2024

  8. [8]

    Kevin Frans, L. B. Soros, and Olaf Witkowski. 2021. CLIPDraw: Exploring Text-to- Drawing Synthesis through Language-Image Encoders. arXiv:2106.14843 [cs.CV] https://arxiv.org/abs/2106.14843

    arXiv 2021

Show all 61 references
  1. [9]

    Jiahui Gao, Renjie Pi, Jipeng Zhang, Jiacheng Ye, Wanjun Zhong, Yufei Wang, Lanqing Hong, Jianhua Han, Hang Xu, Zhenguo Li, et al. 2023. G-llava: Solving geometric problem with multi-modal large language model. arXiv preprint arXiv:2312.11370 (2023)

    2023 arXiv

  2. [10]

    Ziyu Guo, Renrui Zhang, Xiangyang Zhu, Yiwen Tang, Xianzheng Ma, Jiaming Han, Kexin Chen, Peng Gao, Xianzhi Li, Hongsheng Li, et al. 2023. Point-bind & point-llm: Aligning point cloud with multi-modality for 3d understanding, generation, and instruction following. arXiv prepri...

    2023 arXiv

  3. [11]

    David Ha and Douglas Eck. 2017. A Neural Representation of Sketch Drawings. arXiv:1704.03477 [cs.NE] https://arxiv.org/abs/1704.03477

    2017 arXiv

  4. [12]

    Azhar Hadmi, William Puech, Brahim Ait Es Said, and Abdellah Ait Ouahman

  5. [13]

    Yining Hong, Haoyu Zhen, Peihao Chen, Shuhong Zheng, Yilun Du, Zhenfang Chen, and Chuang Gan. 2023. 3d-llm: Injecting the 3d world into large language models. Advances in Neural Information Processing Systems 36 (2023), 20482– 20494

    2023

  6. [14]

    Anwen Hu, Yaya Shi, Haiyang Xu, Jiabo Ye, Qinghao Ye, Ming Yan, Chenliang Li, Qi Qian, Ji Zhang, and Fei Huang. 2024. mplug-paperowl: Scientific diagram analysis with the multimodal large language model. In Proceedings of the 32nd ACM International Conference on Multimedia . 6929–6938

    2024

  7. [15]

    Ajay Jain, Amber Xie, and Pieter Abbeel. 2022. VectorFusion: Text-to-SVG by Abstracting Pixel-Based Diffusion Models. arXiv:2211.11319 [cs.CV] https: //arxiv.org/abs/2211.11319

    2022 arXiv

  8. [16]

    Ajay Jain, Amber Xie, and Pieter Abbeel. 2023. VectorFusion: Text-to-SVG by Abstracting Pixel-Based Diffusion Models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 1911–1920

    2023

  9. [17]

    Yuyang Ji and Haohan Wang. 2025. Socratic Chart: Cooperating Multiple Agents for Robust SVG Chart Understanding. arXiv:2504.09764 [cs.CV] https://arxiv. org/abs/2504.09764

    2025 arXiv

  10. [18]

    Albert Q Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, et al . 2024. Mixtral of experts. arXiv preprint arXiv:2401.04088 (2024)

    2024 arXiv

  11. [19]

    Jacob Devlin Ming-Wei Chang Kenton and Lee Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of naacL-HLT, Vol. 1. Minneapolis, Minnesota, 2

    2019

  12. [20]

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al

  13. [21]

    Jusung Lee, Sungguk Cha, Younghyun Lee, and Cheoljong Yang. 2024. Visual question answering instruction: Unlocking multimodal large language model to domain-specific visual multitasks. arXiv preprint arXiv:2402.08360 (2024)

    2024 arXiv

  14. [22]

    Bo Li, Yuanhan Zhang, Liangyu Chen, Jinghao Wang, Fanyi Pu, Jingkang Yang, Chunyuan Li, and Ziwei Liu. 2023. Mimic-it: Multi-modal in-context instruction tuning. arXiv preprint arXiv:2306.05425 (2023)

    2023 arXiv

  15. [23]

    Feng Li, Renrui Zhang, Hao Zhang, Yuanhan Zhang, Bo Li, Wei Li, Zejun Ma, and Chunyuan Li. 2024. Llava-next-interleave: Tackling multi-image, video, and 3d in large multimodal models. arXiv preprint arXiv:2407.07895 (2024)

    2024 arXiv

  16. [24]

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In International conference on machine learning . PMLR, 12888–12900

    2022

  17. [25]

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. 2023. Improved Baselines with Visual Instruction Tuning

    2023

  18. [26]

    Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. 2024. LLaVA-NeXT: Improved reasoning, OCR, and world knowl- edge. https://llava-vl.github.io/blog/2024-01-30-llava-next/

    2024

  19. [27]

    Raphael Gontijo Lopes, David Ha, Douglas Eck, and Jonathon Shlens. 2019. A Learned Representation for Scalable Vector Graphics. arXiv:1904.02632 [cs.CV] https://arxiv.org/abs/1904.02632

    2019 arXiv

  20. [28]

    Chenyang Lyu, Minghao Wu, Longyue Wang, Xinting Huang, Bingshuai Liu, Zefeng Du, Shuming Shi, and Zhaopeng Tu. 2023. Macaw-llm: Multi-modal language modeling with image, audio, video, and text integration. arXiv preprint arXiv:2306.09093 (2023)

    2023 arXiv

  21. [29]

    Xu Ma, Yuqian Zhou, Xingqian Xu, Bin Sun, Valerii Filev, Nikita Orlov, Yun Fu, and Humphrey Shi. 2022. Towards Layer-Wise Image Vectorization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 16314–16323

    2022

  22. [30]

    Maizhen Ning, Zihao Zhou, Qiufeng Wang, Xiaowei Huang, and Kaizhu Huang

  23. [31]

    Kunato Nishina and Yusuke Matsui. 2024. SVGEditBench: A Benchmark Dataset for Quantitative Assessment of LLM’s SVG Editing Capabilities. arXiv preprint arXiv:2404.13710 (2024)

    2024 arXiv

  24. [32]

    OpenAI. 2023. GPT-4V(ision) System Card. (2023). https://openai.com/index/gpt- 4v-system-card/

    2023

  25. [33]

    Antoine Quint. 2003. Scalable vector graphics. IEEE MultiMedia 10, 3 (2003), 99–102

    2003

  26. [34]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...

    2021

  27. [35]

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, et al. 2024. Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714 (2024)

    2024 arXiv

  28. [36]

    Pradyumna Reddy, Michael Gharbi, Michal Lukac, and Niloy J. Mitra. 2021. Im2Vec: Synthesizing Vector Graphics Without Vector Supervision. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 7342–7351

    2021

  29. [37]

    Machel Reid, Nikolay Savinov, Denis Teplyashin, Dmitry Lepikhin, Timothy Lillicrap, Jean-baptiste Alayrac, Radu Soricut, Angeliki Lazaridou, Orhan Firat, Julian Schrittwieser, et al. 2024. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. arX...

    2024 arXiv

  30. [38]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. arXiv:1908.10084 [cs.CL] https://arxiv.org/abs/ 1908.10084

    2019 arXiv

  31. [39]

    Shuhuai Ren, Linli Yao, Shicheng Li, Xu Sun, and Lu Hou. 2024. Timechat: A time-sensitive multimodal large language model for long video understanding. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 14313–14323

    2024

  32. [40]

    Rodriguez, Abhay Puri, Shubham Agarwal, Issam H

    Juan A. Rodriguez, Abhay Puri, Shubham Agarwal, Issam H. Laradji, Pau Ro- driguez, Sai Rajeswar, David Vazquez, Christopher Pal, and Marco Pedersoli. 2024. StarVector: Generating Scalable Vector Graphics Code from Images and Text. arXiv:2312.11556 [cs.CV] https://arxiv.org/abs...

    2024 arXiv

  33. [41]

    Ashish Shenoy, Yichao Lu, Srihari Jayakumar, Debojeet Chatterjee, Mohsen Moslehpour, Pierce Chuang, Abhay Harpale, Vikas Bhardwaj, Di Xu, Shicong Zhao, et al. 2024. Lumos: Empowering multimodal llms with scene text recogni- tion. In Proceedings of the 30th ACM SIGKDD Conferenc...

    2024

  34. [42]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971 (2023)

    2023 arXiv

  35. [43]

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yas- mine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhos- ale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 (2023)

    2023 arXiv

  36. [44]

    Bovik, H.R

    Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. 2004. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing 13, 4 (2004), 600–612. doi:10.1109/TIP.2003.819861 MM ’25, October 27–31, 2025, Dublin, Ireland. Li et al

    2004

  37. [45]

    Jianzong Wu, Xiangtai Li, Chenyang Si, Shangchen Zhou, Jingkang Yang, Jiangn- ing Zhang, Yining Li, Kai Chen, Yunhai Tong, Ziwei Liu, et al. 2024. Towards language-driven video inpainting via multimodal large language models. In Pro- ceedings of the IEEE/CVF Conference on Comp...

    2024

  38. [46]

    Ronghuan Wu, Wanchao Su, Kede Ma, and Jing Liao. 2023. IconShop: Text-Guided Vector Icon Synthesis with Autoregressive Transformers. ACM Trans. Graph. 42, 6, Article 230 (Dec. 2023), 14 pages. doi:10.1145/3618364

    2023 doi

  39. [47]

    Ximing Xing, Haitao Zhou, Chuang Wang, Jing Zhang, Dong Xu, and Qian Yu. 2024. SVGDreamer: Text guided SVG generation with diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4546–4555

    2024

  40. [48]

    Yiying Yang, Wei Cheng, Sijin Chen, Xianfang Zeng, Fukun Yin, Jiaxu Zhang, Liao Wang, Gang Yu, Xingjun Ma, and Yu-Gang Jiang. 2025. OmniSVG: A Unified Scalable Vector Graphics Generation Model. arXiv:2504.06263 [cs.CV] https://arxiv.org/abs/2504.06263

    2025

  41. [49]

    Renrui Zhang, Xiangfei Hu, Bohao Li, Siyuan Huang, Hanqiu Deng, Yu Qiao, Peng Gao, and Hongsheng Li. 2023. Prompt, generate, then cache: Cascade of foundation models makes strong few-shot learners. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni...

    2023

  42. [50]

    Efros, Eli Shechtman, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang

  43. [51]

    Renrui Zhang, Liuhui Wang, Yu Qiao, Peng Gao, and Hongsheng Li. 2023. Learn- ing 3d representations from 2d pre-trained models via image-to-point masked autoencoders. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 21769–21780

    2023

  44. [52]

    Weinberger, and Yoav Artzi

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. BERTScore: Evaluating Text Generation with BERT. arXiv:1904.09675 [cs.CL] https://arxiv.org/abs/1904.09675

    2020 arXiv

  45. [53]

    Henry Hengyuan Zhao, Pan Zhou, Difei Gao, Zechen Bai, and Mike Zheng Shou

  46. [54]

    Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mohamed Elhoseiny. 2023. Minigpt-4: Enhancing vision-language understanding with advanced large lan- guage models. arXiv preprint arXiv:2304.10592 (2023)

    2023 arXiv

  47. [55]

    Bocheng Zou, Mu Cai, Jianrui Zhang, and Yong Jae Lee. 2024. VGBench: Evaluat- ing Large Language Models on Vector Graphics Understanding and Generation. arXiv preprint arXiv:2407.10972 (2024)

    2024 arXiv

  48. [2012]

    In Watermarking, Mithun Das Gupta (Ed.)

    Perceptual Image Hashing. In Watermarking, Mithun Das Gupta (Ed.). IntechOpen, Rijeka, Chapter 2. doi:10.5772/37435

    doi

  49. [2018]

    arXiv:1801.03924 [cs.CV] https://arxiv.org/abs/1801.03924

    The Unreasonable Effectiveness of Deep Features as a Perceptual Metric. arXiv:1801.03924 [cs.CV] https://arxiv.org/abs/1801.03924

    arXiv

  50. [2022]

    Advances in neural information processing systems 35 (2022), 23716–23736

    Flamingo: a visual language model for few-shot learning. Advances in neural information processing systems 35 (2022), 23716–23736

    2022

  51. [2023]

    In Proceedings of the IEEE/CVF International Conference on Computer Vision

    Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 4015–4026

  52. [2024]

    In Advances in Neural Information Processing Systems , A

    LOVA3: Learning to Visual Question Answering, Asking and Assessment. In Advances in Neural Information Processing Systems , A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang (Eds.), Vol. 37. Curran Associates, Inc., 115146–115175. https://procee...

    2024

  53. [2025]

    In Proceedings of the AAAI Conference on Artificial Intelligence, Vol

    GNS: Solving Plane Geometry Problems by Neural-Symbolic Reasoning with Multi-Modal LLMs. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 24957–24965

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.