Pith. sign in

REVIEW 3 major objections 1 minor 1 cited by

CRISP: Contrastive Residual Injection and Semantic Prompting for Continual Video Instance Segmentation

T0 review · 3 major / 1 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 CRISP claims a video segmentation method that learns new classes without forgetting old ones, yet the supplied body text is a different paper.

desk verdict The abstract and full text are two different papers; CRISP cannot be evaluated from this submission. read the letter →

arxiv 2508.10432 v1 pith:I4HAJI2R submitted 2025-08-14 cs.CV

Baichen Liu , Qi Lyu , Xudong Wang , Jiahua Dong , Lianqing Liu , Zhi Han This is my paper · ORCID
classification cs.CV
keywords continualvideoinstancesegmentationcatastrophicforgettingcontrastivelearningsemanticpromptingadaptiveresidualpromptincrementalYouTube-VIScorrelationloss
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 CRISP (Contrastive Residual Injection and Semantic Prompting), a method for continual video instance segmentation that aims to learn new object categories without forgetting previously learned ones. The abstract claims CRISP significantly outperforms existing continual segmentation methods on YouTube-VIS-2019 and YouTube-VIS-2021, resolving instance-wise, category-wise, and task-wise confusion. The claimed mechanism pairs an instance correlation loss with an adaptive residual semantic prompt (ARSP) pool generated from category text, plus a contrastive semantic consistency loss and an incremental prompt initialization strategy. The full text supplied for this review is actually a mathematics-reasoning paper (WE-MATH 2.0), so none of CRISP's equations, ablations, or experimental tables appears in the body. Thus the abstract's claims are on record, but the document as provided cannot support or refute them.

What carries the argument

The load-bearing components are the three training-time mechanisms: the instance correlation loss (which treats the prior query space as an anchor and enforces current-task specificity), the adaptive residual semantic prompt (ARSP) pool (category text projected into learnable residual prompts and matched to object queries), and the contrastive semantic consistency loss (which pulls object queries and residual prompts into semantic coherence). An incremental prompt initialization strategy is the fourth component, meant to preserve inter-task query-space correlations. Together these are claimed to prevent forgetting at the instance, category, and task levels.

What would settle it

Look inside the manuscript for CRISP's method section: if neither ARSP, the instance correlation loss, the semantic consistency loss, nor the YouTube-VIS-2019/2021 experimental tables appear anywhere, the central claim has no support in the submitted document. Alternatively, run CRISP's public code on YouTube-VIS with disjoint category tasks and compare old-class AP after each task to a standard fine-tuned baseline; a non-improvement there would refute the forgetting-avoidance claim.

Watch

Extended reading notes

Core claim

The central claim, as stated in the abstract, is that CRISP outperforms existing continual segmentation methods on long-term continual video instance segmentation on YouTube-VIS-2019 and YouTube-VIS-2021 while avoiding catastrophic forgetting. Its designed contributions are: (1) an instance correlation loss that models tracking by aligning current task queries with the prior query space while sharpening current-task specificity; (2) ARSP, a learnable residual prompt pool generated from category text, with an adjustive query-prompt matching mechanism; (3) a contrastive semantic consistency loss linking object queries and residual prompts during incremental training; and (4) a prompt initializ

Load-bearing premise

The load-bearing premise is document-level: the body text must actually describe CRISP and its experiments for the claimed state-of-the-art result to be checkable; here the body is an unrelated mathematics-reasoning paper, so the abstract's claim stands unsupported by the supplied text.

Editorial extensions

If this is right

  • If CRISP works as claimed, continual video instance segmentation models can add new categories across tasks without a steep drop in old-category performance.
  • The ARSP query-prompt matching implies new categories can be incorporated through prompt-pool assignment rather than full model retraining.
  • The contrastive semantic consistency loss should keep object queries semantically stable as tasks accumulate, directly attacking catastrophic forgetting.
  • On the reported benchmarks, CRISP would set the current state of the art among continual segmentation methods for long-term settings.
  • The dual losses and prompt initialization target the three distinct confusion types, so each component is a separable intervention for diagnosis.

Reading between the lines

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

  • My reading: the document-level mismatch means the substantive contribution of this submission cannot be evaluated from the provided text; if the WE-MATH 2.0 body was attached by retrieval error, the CRISP evaluation requires the original manuscript.
  • If the abstract's design is taken at face value, the ARSP pool offers a transferable pattern: category text as a prior for residual prompts could apply to other incremental recognition settings, such as open-vocabulary detection or long-tail classification, not just video instance segmentation.
  • A testable extension would be to ablate the three losses separately against single-task continual baselines; the abstract does not report which loss carries the main forgetting reduction.
  • The claimed 'concise yet powerful' prompt initialization strategy would be worth testing as a standalone replay-free baseline for task-wise forgetting.
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 / 1 minor

Summary. The submission is headed arXiv:2508.10432 and titled 'CRISP: Contrastive Residual Injection and Semantic Prompting for Continual Video Instance Segmentation'. The abstract promises a method (CRISP) with three components—an instance correlation loss, an adaptive residual semantic prompt (ARSP) learning framework with query-prompt matching, a contrastive semantic consistency loss, and an incremental prompt initialization strategy—and claims state-of-the-art performance on YouTube-VIS-2019 and YouTube-VIS-2021. However, the full text supplied for review is a different paper: 'WE-MATH 2.0: A Versatile MathBook System for Incentivizing Visual Mathematical Reasoning' (arXiv:2508.10433v1 [cs.AI]). That full text is about multimodal mathematical reasoning and never mentions CRISP, continual video instance segmentation, ARSP, the three claimed losses, or YouTube-VIS. The submitted document therefore contains no architecture, no equations, no experiments, and no ablations supporting the abstract's claims.

Significance. If CRISP were fully specified and its experimental claims supported, it would be a credible contribution to continual video instance segmentation, a task of active interest in the computer vision community. The claimed combination of instance-, category-, and task-wise mechanisms is potentially interesting. However, the manuscript under review contains none of the required evidence: there is no method section for CRISP, no definition of the losses, no definition of the ARSP pool, and no experiments on the claimed benchmarks. The abstract's GitHub link cannot compensate for the absence of the method and results in the body. Consequently, the significance of the work cannot be assessed from this document; there is also no machine-checked proof or reproducible code for CRISP to credit, since the body does not mention CRISP at all.

major comments (3)
  1. [Abstract vs. Full Text (document-level mismatch)] The full text is not the CRISP paper. Its title is 'WE-MATH 2.0: A Versatile MathBook System for Incentivizing Visual Mathematical Reasoning' and its header reads arXiv:2508.10433v1 [cs.AI], not arXiv:2508.10432. The body never mentions CRISP, ARSP, instance correlation loss, semantic consistency loss, continual video instance segmentation, or YouTube-VIS. Under the in-scope evidence rule, this full text is the only evidence in the manuscript. The abstract's central claim — that CRISP significantly outperforms existing continual segmentation methods on YouTube-VIS-2019 and YouTube-VIS-2021 — is therefore entirely unsupported by the submitted document.
  2. [§1 / Method (CRISP components)] The abstract lists three load-bearing design components: instance correlation loss, adaptive residual semantic prompt (ARSP) learning with an adjustive query-prompt matching mechanism, and a semantic consistency loss based on contrastive learning, plus an initialization strategy for incremental prompts. None of these components is defined, formulated, or even named anywhere in the full text. There is no equation for any loss, no definition of the prompt pool or its learning rule, and no description of the matching mechanism. The method is therefore unassessable; there is no way to check for internal consistency, novelty, or correctness.
  3. [§5 / Experiments] The claimed experiments on YouTube-VIS-2019 and YouTube-VIS-2021 do not appear in the submitted document. The full text reports experiments on MathVista, MathVision, We-Math, MathVerse, and MathBookEval for the WE-MATH 2.0 system. There are no continual video instance segmentation baselines, no evaluation metrics for VIS, no comparison tables involving CRISP, and no ablation of the three claimed losses or of the prompt initialization strategy. Thus the abstract's empirical claim of 'significantly outperforming existing continual segmentation methods' has no supporting data in the manuscript.
minor comments (1)
  1. [Title page and metadata] The title, author list, and arXiv identifier on the header do not match the abstract. The reference list likewise corresponds to the WE-MATH 2.0 paper, not to CRISP. Even the abstract's GitHub URL for CRISP cannot be cross-checked against the body because the body contains no code availability statement for CRISP.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity can be identified because the supplied full text is arXiv:2508.10433 (WE-MATH 2.0), not the CRISP paper; the CRISP method, equations, and experiments are absent, so there is no derivation chain to audit.

full rationale

The claimed paper is CRISP (arXiv:2508.10432, cs.CV), whose abstract describes Contrastive Residual Injection and Semantic Prompting with ARSP, three losses, and experiments on YouTube-VIS-2019/2021. The supplied full text, however, is arXiv:2508.10433, 'WE-MATH 2.0: A Versatile MathBook System for Incentivizing Visual Mathematical Reasoning' (cs.AI). The body text introduces a MathBook knowledge hierarchy, GeoGebra-rendered datasets, a two-stage RL pipeline, and evaluations on MathVista, MathVision, We-Math, and MathVerse. It never mentions CRISP, ARSP, video instance segmentation, continual segmentation, YouTube-VIS, instance correlation loss, semantic consistency loss, or prompt initialization. Under the hard rules, circularity may only be claimed when a specific reduction can be quoted and exhibited (e.g., Eq. X = Eq. Y by construction, or a fitted parameter renamed as a prediction). No such reduction is present for CRISP because none of CRISP's components appear in the document. The document-level body/abstract mismatch is a severe correctness and provenance issue, but it is not evidence of circular reasoning: there is no method section whose derivations could reduce to their inputs. Accordingly, the honest circularity finding is no significant circularity identified (score 0), with the caveat that the CRISP content is entirely missing from this review and therefore the non-circularity of its claims is not affirmatively established.

Assumptions & free parameters 0 free parameters · 1 assumptions · 1 invented entities

The ledger is nearly empty because the CRISP content is absent. The only identifiable additions are the ARSP construct from the abstract, which has no visible implementation or independent falsifiable handle, and the document-correspondence assumption, which fails.

assumptions (1)
  • ad hoc to paper The body text corresponds to the paper named in the title and abstract (CRISP, arXiv 2508.10432)
    This assumption is required for any technical evaluation, and it fails: the body is the title page and full text of WE-MATH 2.0 (arXiv 2508.10433), an unrelated paper about multimodal math reasoning.
invented entities (1)
  • Adaptive Residual Semantic Prompt (ARSP) pool
    purpose: Claimed to build a mapping between current-task queries and category-text-derived residual prompts for category-wise continual learning
    Mentioned only in the abstract; no architecture, equations, or experiments appear in the provided full text, so it has no falsifiable handle outside this document.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CRISP: Contrastive Residual Injection and Semantic Prompting for Continual Video Instance Segmentation." pith.science (2026). https://pith.science/paper/I4HAJI2R

@misc{pith2026250810432,
  author       = {Pith},
  title        = {Pith review of: CRISP: Contrastive Residual Injection and Semantic Prompting for Continual Video Instance Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I4HAJI2R}},
  note         = {Machine review of arXiv:2508.10432}
}
read the original abstract

Continual video instance segmentation demands both the plasticity to absorb new object categories and the stability to retain previously learned ones, all while preserving temporal consistency across frames. In this work, we introduce Contrastive Residual Injection and Semantic Prompting (CRISP), an earlier attempt tailored to address the instance-wise, category-wise, and task-wise confusion in continual video instance segmentation. For instance-wise learning, we model instance tracking and construct instance correlation loss, which emphasizes the correlation with the prior query space while strengthening the specificity of the current task query. For category-wise learning, we build an adaptive residual semantic prompt (ARSP) learning framework, which constructs a learnable semantic residual prompt pool generated by category text and uses an adjustive query-prompt matching mechanism to build a mapping relationship between the query of the current task and the semantic residual prompt. Meanwhile, a semantic consistency loss based on the contrastive learning is introduced to maintain semantic coherence between object queries and residual prompts during incremental training. For task-wise learning, to ensure the correlation at the inter-task level within the query space, we introduce a concise yet powerful initialization strategy for incremental prompts. Extensive experiments on YouTube-VIS-2019 and YouTube-VIS-2021 datasets demonstrate that CRISP significantly outperforms existing continual segmentation methods in the long-term continual video instance segmentation task, avoiding catastrophic forgetting and effectively improving segmentation and classification performance. The code is available at https://github.com/01upup10/CRISP.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Interleaved Transceiver Design for a Continuous- Transmission MIMO-OFDM ISAC System

    eess.SP 2025-08 unverdicted novelty 4.0 of 10

    The abstract advertises a MIMO-OFDM ISAC transceiver design with a claimed first ADPM convergence proof, but the full text is a different paper on continual video instance segmentation, making the ISAC claims unreviewable.

Reference graph

Works this paper leans on

77 extracted references · 71 canonical work pages · cited by 1 Pith paper

  1. [1]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. ����� �������� ����������������, 2023

    work page 2023

  2. [2]

    M2-reasoning: Empowering mllms with unified general and spatial reasoning

    Inclusion AI, Fudong Wang, Jiajia Liu, Jingdong Chen, Jun Zhou, Kaixiang Ji, Lixiang Ru, Qingpei Guo, Ruobing Zheng, Tianqi Li, et al. M2-reasoning: Empowering mllms with unified general and spatial reasoning. ����� �������� ����������������, 2025

    work page 2025

  3. [3]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report. ����� �������� ����������������, 2025

    work page 2025

  4. [4]

    Qwen2.5-vl technical report

    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. Qwen2.5-vl technical report. �...

    2025

  5. [5]

    Sft or rl? an early investigation into training r1-like reasoning large vision-language models

    Hardy Chen, Haoqin Tu, Fali Wang, Hui Liu, Xianfeng Tang, Xinya Du, Yuyin Zhou, and Cihang Xie. Sft or rl? an early investigation into training r1-like reasoning large vision-language models. ����� �������� ����������������, 2025

    work page 2025

  6. [6]

    Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling

    Zhe Chen, Weiyun Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Erfei Cui, Jinguo Zhu, Shenglong Ye, Hao Tian, Zhaoyang Liu, Lixin Gu, Xuehui Wang, Qingyun Li, Yimin Ren, Zixuan Chen, Jiapeng Luo, Jiahao Wang, Tan Jiang, Bo Wang, Conghui He, Botian Shi, Xingcheng Zhang, Han Lv, Yi Wang, Wenqi Shao, Pei Chu, Zhongying Tu, Tong He, Zhiyong Wu, Huipeng Deng, Jia...

    arXiv 2024

  7. [7]

    How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites

    Zhe Chen, Weiyun Wang, Hao Tian, Shenglong Ye, Zhangwei Gao, Erfei Cui, Wenwen Tong, Kongzhi Hu, Jiapeng Luo, Zheng Ma, et al. How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites. ����� �������� ����������������, 2024

    work page 2024

  8. [8]

    Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks

    Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, Bin Li, Ping Luo, Tong Lu, Yu Qiao, and Jifeng Dai. Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks. ����� �������� ����������������, 2023

    work page 2023

Show all 77 references
  1. [9]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning, 2025

    DeepSeek-AI. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning, 2025

    2025

  2. [10]

    Open- vlthinker: An early exploration to complex vision-language reasoning via iterative self- improvement

    Yihe Deng, Hritik Bansal, Fan Yin, Nanyun Peng, Wei Wang, and Kai-Wei Chang. Open- vlthinker: An early exploration to complex vision-language reasoning via iterative self- improvement. ����� �������� ����������������, 2025

    2025

  3. [11]

    Tool-star: Empowering llm-brained multi-tool reasoner via reinforcement learning

    Guanting Dong, Yifei Chen, Xiaoxi Li, Jiajie Jin, Hongjin Qian, Yutao Zhu, Hangyu Mao, Guorui Zhou, Zhicheng Dou, and Ji-Rong Wen. Tool-star: Empowering llm-brained multi-tool reasoner via reinforcement learning. ����� �������� ����������������, 2025

    2025

  4. [12]

    Agentic reinforced policy optimization

    Guanting Dong, Hangyu Mao, Kai Ma, Licheng Bao, Yifei Chen, Zhongyuan Wang, Zhongxia Chen, Jiazhen Du, Huiyang Wang, Fuzheng Zhang, et al. Agentic reinforced policy optimization. ����� �������� ����������������, 2025

    2025

  5. [13]

    Progressive multimodal reasoning via active retrieval

    Guanting Dong, Chenghao Zhang, Mengjie Deng, Yutao Zhu, Zhicheng Dou, and Ji-Rong Wen. Progressive multimodal reasoning via active retrieval. ����� �������� ����������������, 2024

    2024

  6. [14]

    Vlmevalkit: An open-source toolkit for evaluating large multi-modality models

    Haodong Duan, Junming Yang, Yuxuan Qiao, Xinyu Fang, Lin Chen, Yuan Liu, Xiaoyi Dong, Yuhang Zang, Pan Zhang, Jiaqi Wang, et al. Vlmevalkit: An open-source toolkit for evaluating large multi-modality models. In ����������� �� ��� ���� ��� ������������� ���������� �� ����������...

    2024

  7. [15]

    Euclid’s elements of geometry

    Richard Fitzpatrick. Euclid’s elements of geometry. 2008

    2008

  8. [16]

    G-llava: Solving geometric problem with multi-modal large language model

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

    2023

  9. [17]

    Chatglm: A family of large language models from glm-130b to glm-4 all tools

    Team GLM, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, et al. Chatglm: A family of large language models from glm-130b to glm-4 all tools. ����� �������� ����������������, 2024

    2024

  10. [18]

    Dong Guo, Faming Wu, Feida Zhu, Fuxing Leng, Guang Shi, Haobin Chen, Haoqi Fan, Jian Wang, Jianyu Jiang, Jiawei Wang, et al. Seed1. 5-vl technical report. ����� �������� ����������������, 2025

    2025

  11. [19]

    Infimm-webmath-40b: Advancing multimodal pre-training for enhanced mathematical reasoning

    Xiaotian Han, Yiren Jian, Xuefeng Hu, Haogeng Liu, Yiqi Wang, Qihang Fan, Yuang Ai, Huaibo Huang, Ran He, Zhenheng Yang, et al. Infimm-webmath-40b: Advancing multimodal pre-training for enhanced mathematical reasoning. ����� �������� ����������������, 2024

    2024

  12. [20]

    Glm-4.1 v-thinking: Towards versatile multimodal reasoning with scalable reinforcement learning

    Wenyi Hong, Wenmeng Yu, Xiaotao Gu, Guo Wang, Guobing Gan, Haomiao Tang, Jiale Cheng, Ji Qi, Junhui Ji, Lihang Pan, et al. Glm-4.1 v-thinking: Towards versatile multimodal reasoning with scalable reinforcement learning. ����� �������� ����������������, 2025

    2025

  13. [21]

    Vision-r1: Incentivizing reasoning capability in multimodal large language models

    Wenxuan Huang, Bohan Jia, Zijie Zhai, Shaosheng Cao, Zheyu Ye, Fei Zhao, Zhe Xu, Yao Hu, and Shaohui Lin. Vision-r1: Incentivizing reasoning capability in multimodal large language models. ����� �������� ����������������, 2025

    2025

  14. [22]

    Openai o1 system card

    Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, et al. Openai o1 system card. ����� �������� ����������������, 2024

    2024

  15. [23]

    Clevr: A diagnostic dataset for compositional language and elementary visual reasoning

    Justin Johnson, Bharath Hariharan, Laurens Van Der Maaten, Li Fei-Fei, C Lawrence Zitnick, and Ross Girshick. Clevr: A diagnostic dataset for compositional language and elementary visual reasoning. In ����������� �� ��� ���� ���������� �� �������� ������ ��� ������� ����������...

    2017

  16. [24]

    Step- dpo: Step-wise preference optimization for long-chain reasoning of llms

    Xin Lai, Zhuotao Tian, Yukang Chen, Senqiao Yang, Xiangru Peng, and Jiaya Jia. Step- dpo: Step-wise preference optimization for long-chain reasoning of llms. ����� �������� ����������������, 2024

    2024

  17. [25]

    Instructerc: Reforming emotion recognition in conversation with multi-task retrieval-augmented large language models

    Shanglin Lei, Guanting Dong, Xiaoping Wang, Keheng Wang, Runqi Qiao, and Sirui Wang. Instructerc: Reforming emotion recognition in conversation with multi-task retrieval-augmented large language models. ����� �������� ����������������, 2023

    2023

  18. [26]

    Gaokao-eval: Does high scores truly reflect strong capabilities in llms? ����� �������� ����������������, 2024

    Zhikai Lei, Tianyi Liang, Hanglei Hu, Jin Zhang, Yunhua Zhou, Yunfan Shao, Linyang Li, Chenchui Li, Changbo Wang, Hang Yan, et al. Gaokao-eval: Does high scores truly reflect strong capabilities in llms? ����� �������� ����������������, 2024

    2024

  19. [27]

    Llava-onevision: Easy visual task transfer

    Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, et al. Llava-onevision: Easy visual task transfer. ����� �������� ����������������, 2024

    2024

  20. [28]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. In �������� ���������� �� �������� ������, pages 38–55. Springer, 2024

    2024

  21. [29]

    Noisyrollout: Reinforcing visual reasoning with data augmentation

    Xiangyan Liu, Jinjie Ni, Zijian Wu, Chao Du, Longxu Dou, Haonan Wang, Tianyu Pang, and Michael Qizhe Shieh. Noisyrollout: Reinforcing visual reasoning with data augmentation. ����� �������� ����������������, 2025

    2025

  22. [30]

    Textmonkey: An ocr-free large multimodal model for understanding document

    Yuliang Liu, Biao Yang, Qiang Liu, Zhang Li, Zhiyin Ma, Shuo Zhang, and Xiang Bai. Textmonkey: An ocr-free large multimodal model for understanding document. ����� �������� ����������������, 2024. 12

    2024

  23. [31]

    Visual-rft: Visual reinforcement fine-tuning

    Ziyu Liu, Zeyi Sun, Yuhang Zang, Xiaoyi Dong, Yuhang Cao, Haodong Duan, Dahua Lin, and Jiaqi Wang. Visual-rft: Visual reinforcement fine-tuning. ����� �������� ����������������, 2025

    2025

  24. [32]

    Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts

    Pan Lu, Hritik Bansal, Tony Xia, Jiacheng Liu, Chunyuan Li, Hannaneh Hajishirzi, Hao Cheng, Kai-Wei Chang, Michel Galley, and Jianfeng Gao. Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts. ����� �������� ����������������, 2023

    2023

  25. [33]

    Inter-gps: Interpretable geometry problem solving with formal language and symbolic reasoning

    Pan Lu, Ran Gong, Shibiao Jiang, Liang Qiu, Siyuan Huang, Xiaodan Liang, and Song-Chun Zhu. Inter-gps: Interpretable geometry problem solving with formal language and symbolic reasoning. ����� �������� ����������������, 2021

    2021

  26. [34]

    Iconqa: A new benchmark for abstract diagram understanding and visual language reasoning

    Pan Lu, Liang Qiu, Jiaqi Chen, Tony Xia, Yizhou Zhao, Wei Zhang, Zhou Yu, Xiaodan Liang, and Song-Chun Zhu. Iconqa: A new benchmark for abstract diagram understanding and visual language reasoning. ����� �������� ����������������, 2021

    2021

  27. [35]

    Ursa: Understanding and verifying chain-of-thought reasoning in multimodal mathematics

    Ruilin Luo, Zhuofan Zheng, Yifan Wang, Yiyao Yu, Xinzhe Ni, Zicheng Lin, Jin Zeng, and Yujiu Yang. Ursa: Understanding and verifying chain-of-thought reasoning in multimodal mathematics. ����� �������� ����������������, 2025

    2025

  28. [36]

    Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforcement learning

    Fanqing Meng, Lingxiao Du, Zongkai Liu, Zhixiang Zhou, Quanfeng Lu, Daocheng Fu, Tiancheng Han, Botian Shi, Wenhai Wang, Junjun He, et al. Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforcement learning. ����� �������� ����������������, 2025

    2025

  29. [37]

    Hello gpt-4o, 2024

    OpenAI. Hello gpt-4o, 2024

    2024

  30. [38]

    Gpt-4v (ision) system card

    R OpenAI. Gpt-4v (ision) system card. �������� ���������, 2023

    2023

  31. [39]

    Multi- math: Bridging visual and mathematical reasoning for large language models

    Shuai Peng, Di Fu, Liangcai Gao, Xiuqin Zhong, Hongguang Fu, and Zhi Tang. Multi- math: Bridging visual and mathematical reasoning for large language models. ����� �������� ����������������, 2024

    2024

  32. [40]

    Toolrl: Reward is all tool learning needs

    Cheng Qian, Emre Can Acikgoz, Qi He, Hongru Wang, Xiusi Chen, Dilek Hakkani-Tür, Gokhan Tur, and Heng Ji. Toolrl: Reward is all tool learning needs. ����� �������� ����������������, 2025

    2025

  33. [41]

    V-oracle: Making progressive reasoning in deciphering oracle bones for you and me

    Runqi Qiao, Qiuna Tan, Guanting Dong, MinhuiWu MinhuiWu, Jiapeng Wang, YiFan Zhang, Zhuoma GongQue, Chong Sun, Yida Xu, Yadong Xue, Ye Tian, Zhimin Bao, Lan Yang, Chen Li, and Honggang Zhang. V-oracle: Making progressive reasoning in deciphering oracle bones for you and me. In...

    2025

  34. [42]

    We-math: Does your large multimodal model achieve human-like mathematical reasoning? ����� �������� ����������������, 2024

    Runqi Qiao, Qiuna Tan, Guanting Dong, Minhui Wu, Chong Sun, Xiaoshuai Song, Zhuoma GongQue, Shanglin Lei, Zhe Wei, Miaoxuan Zhang, et al. We-math: Does your large multimodal model achieve human-like mathematical reasoning? ����� �������� ����������������, 2024

    2024

  35. [43]

    We-math: Does your large multimodal model achieve human-like mathematical reasoning? ���� , abs/2407.01284, 2024

    Runqi Qiao, Qiuna Tan, Guanting Dong, Minhui Wu, Chong Sun, Xiaoshuai Song, Zhuoma Gongque, Shanglin Lei, Zhe Wei, Miaoxuan Zhang, Runfeng Qiao, Yifan Zhang, Xiao Zong, Yida Xu, Muxi Diao, Zhimin Bao, Chen Li, and Honggang Zhang. We-math: Does your large multimodal model achie...

    2024 arXiv

  36. [44]

    Making visual sense of oracle bones for you and me

    Runqi Qiao, Lan Yang, Kaiyue Pang, and Honggang Zhang. Making visual sense of oracle bones for you and me. In ����������� �� ��� �������� ���������� �� �������� ������ ��� ������� ����������� ������, pages 12656–12665, June 2024

    2024

  37. [45]

    Grounding dino 1.5: Advance the" edge" of open-set object detection

    Tianhe Ren, Qing Jiang, Shilong Liu, Zhaoyang Zeng, Wenlong Liu, Han Gao, Hongjie Huang, Zhengyu Ma, Xiaoke Jiang, Yihao Chen, et al. Grounding dino 1.5: Advance the" edge" of open-set object detection. ����� �������� ����������������, 2024. 13

    2024

  38. [46]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. ����� �������� ����������������, 2024

    2024

  39. [47]

    Math-LLaV A: Bootstrapping mathematical reasoning for multimodal large language models

    Wenhao Shi, Zhiqiang Hu, Yi Bin, Junhua Liu, Yang Yang, See-Kiong Ng, Lidong Bing, and Roy Ka-Wei Lee. Math-LLaV A: Bootstrapping mathematical reasoning for multimodal large language models. In �������� �� ��� ����������� ��� ������������� ������������ ����� ����, pages 4663–4...

    2024

  40. [48]

    Math-llava: Bootstrapping mathematical reasoning for multimodal large language models

    Wenhao Shi, Zhiqiang Hu, Yi Bin, Junhua Liu, Yang Yang, See-Kiong Ng, Lidong Bing, and Roy Ka-Wei Lee. Math-llava: Bootstrapping mathematical reasoning for multimodal large language models. ����� �������� ����������������, 2024

    2024

  41. [49]

    Cs-bench: A comprehensive benchmark for large language models towards computer science mastery

    Xiaoshuai Song, Muxi Diao, Guanting Dong, Zhengyang Wang, Yujia Fu, Runqi Qiao, Zhexu Wang, Dayuan Fu, Huangxuan Wu, Bin Liang, et al. Cs-bench: A comprehensive benchmark for large language models towards computer science mastery. ����� �������� ����������������, 2024

    2024

  42. [50]

    Mimo-vl technical report, 2025

    Core Team, Zihao Yue, Zhenru Lin, Yifan Song, Weikun Wang, Shuhuai Ren, Shuhao Gu, Shicheng Li, Peidian Li, Liang Zhao, Lei Li, Kainan Bao, Hao Tian, Hailin Zhang, Gang Wang, Dawei Zhu, Cici, Chenhong He, Bowen Ye, Bowen Shen, Zihan Zhang, Zihan Jiang, Zhixian Zheng, Zhichao S...

    2025

  43. [51]

    Gemini: a family of highly capable multimodal models

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. Gemini: a family of highly capable multimodal models. ����� �������� ����������������, 2023

    2023

  44. [52]

    Kwai keye-vl technical report

    Kwai Keye Team, Biao Yang, Bin Wen, Changyi Liu, Chenglong Chu, Chengru Song, Chongling Rao, Chuan Yi, Da Li, Dunju Zang, et al. Kwai keye-vl technical report. ����� �������� ����������������, 2025

    2025

  45. [53]

    Solving olympiad geometry without human demonstrations

    Trieu H Trinh, Yuhuai Wu, Quoc V Le, He He, and Thang Luong. Solving olympiad geometry without human demonstrations. ������, 625(7995):476–482, 2024

    2024

  46. [54]

    Solving math word problems with process-and outcome-based feedback

    Jonathan Uesato, Nate Kushman, Ramana Kumar, Francis Song, Noah Siegel, Lisa Wang, Antonia Creswell, Geoffrey Irving, and Irina Higgins. Solving math word problems with process-and outcome-based feedback. ����� �������� ����������������, 2022

    2022

  47. [55]

    Srpo: Enhancing multimodal llm reasoning via reflection- aware reinforcement learning

    Zhongwei Wan, Zhihao Dou, Che Liu, Yu Zhang, Dongfei Cui, Qinjian Zhao, Hui Shen, Jing Xiong, Yi Xin, Yifan Jiang, et al. Srpo: Enhancing multimodal llm reasoning via reflection- aware reinforcement learning. ����� �������� ����������������, 2025

    2025

  48. [56]

    Alphazero-like tree-search can guide large language model decoding and training

    Ziyu Wan, Xidong Feng, Muning Wen, Stephen Marcus McAleer, Ying Wen, Weinan Zhang, and Jun Wang. Alphazero-like tree-search can guide large language model decoding and training. In ���������� ������������� ���������� �� ������� ��������, 2024

    2024

  49. [57]

    Measuring multimodal mathematical reasoning with math-vision dataset

    Ke Wang, Junting Pan, Weikang Shi, Zimu Lu, Houxing Ren, Aojun Zhou, Mingjie Zhan, and Hongsheng Li. Measuring multimodal mathematical reasoning with math-vision dataset. �������� �� ������ ����������� ���������� �������, 37:95095–95169, 2024

    2024

  50. [58]

    Measuring multimodal mathematical reasoning with math-vision dataset

    Ke Wang, Junting Pan, Weikang Shi, Zimu Lu, Houxing Ren, Aojun Zhou, Mingjie Zhan, and Hongsheng Li. Measuring multimodal mathematical reasoning with math-vision dataset. �������� �� ������ ����������� ���������� �������, 37:95095–95169, 2025

    2025

  51. [59]

    Mathcoder-vl: Bridging vision and code for enhanced multimodal mathematical reasoning

    Ke Wang, Junting Pan, Linda Wei, Aojun Zhou, Weikang Shi, Zimu Lu, Han Xiao, Yunqiao Yang, Houxing Ren, Mingjie Zhan, et al. Mathcoder-vl: Bridging vision and code for enhanced multimodal mathematical reasoning. ����� �������� ����������������, 2025. 14

    2025

  52. [60]

    Math-shepherd: A label-free step-by-step verifier for llms in mathematical reasoning

    Peiyi Wang, Lei Li, Zhihong Shao, RX Xu, Damai Dai, Yifei Li, Deli Chen, Y Wu, and Zhifang Sui. Math-shepherd: A label-free step-by-step verifier for llms in mathematical reasoning. ����� �������� ����������������, 2023

    2023

  53. [61]

    General ocr theory: Towards ocr-2.0 via a unified end-to-end model

    Haoran Wei, Chenglong Liu, Jinyue Chen, Jia Wang, Lingyu Kong, Yanming Xu, Zheng Ge, Liang Zhao, Jianjian Sun, Yuang Peng, et al. General ocr theory: Towards ocr-2.0 via a unified end-to-end model. 2024

    2024

  54. [62]

    Chi, Quoc V

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V . Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models. In Sanmi Koyejo, S. Mohamed, A. Agarwal, Danielle Belgrave, K. Cho, and A. Oh, editors,...

    2022

  55. [63]

    Deepseek-prover-v1

    Huajian Xin, ZZ Ren, Junxiao Song, Zhihong Shao, Wanjia Zhao, Haocheng Wang, Bo Liu, Liyue Zhang, Xuan Lu, Qiushi Du, et al. Deepseek-prover-v1. 5: Harnessing proof assistant feed- back for reinforcement learning and monte-carlo tree search. ����� �������� ����������������, 2024

    2024

  56. [64]

    Geosense: Evaluating identification and application of geometric principles in multimodal reasoning

    Liangyu Xu, Yingxiu Zhao, Jingyun Wang, Yingyao Wang, Bu Pi, Chen Wang, Mingliang Zhang, Jihao Gu, Xiang Li, Xiaoyong Zhu, et al. Geosense: Evaluating identification and application of geometric principles in multimodal reasoning. ����� �������� ����������������, 2025

    2025

  57. [65]

    Wethink: Toward general-purpose vision-language reasoning via reinforcement learning

    Jie Yang, Feipeng Ma, Zitian Wang, Dacheng Yin, Kang Rong, Fengyun Rao, and Ruimao Zhang. Wethink: Toward general-purpose vision-language reasoning via reinforcement learning. ����� �������� ����������������, 2025

    2025

  58. [66]

    R1-onevision: Advancing generalized multimodal reasoning through cross-modal formalization

    Yi Yang, Xiaoxuan He, Hongkun Pan, Xiyan Jiang, Yan Deng, Xingtao Yang, Haoyu Lu, Dacheng Yin, Fengyun Rao, Minfeng Zhu, et al. R1-onevision: Advancing generalized multimodal reasoning through cross-modal formalization. ����� �������� ����������������, 2025

    2025

  59. [67]

    mplug-docowl: Modularized multimodal large language model for document understanding

    Jiabo Ye, Anwen Hu, Haiyang Xu, Qinghao Ye, Ming Yan, Yuhao Dan, Chenlin Zhao, Guohai Xu, Chenliang Li, Junfeng Tian, et al. mplug-docowl: Modularized multimodal large language model for document understanding. ����� �������� ����������������, 2023

    2023

  60. [68]

    Ureader: Universal ocr-free visually-situated language understanding with multimodal large language model

    Jiabo Ye, Anwen Hu, Haiyang Xu, Qinghao Ye, Ming Yan, Guohai Xu, Chenliang Li, Jun- feng Tian, Qi Qian, Ji Zhang, et al. Ureader: Universal ocr-free visually-situated language understanding with multimodal large language model. ����� �������� ����������������, 2023

    2023

  61. [69]

    Limo: Less is more for reasoning

    Yixin Ye, Zhen Huang, Yang Xiao, Ethan Chern, Shijie Xia, and Pengfei Liu. Limo: Less is more for reasoning. ����� �������� ����������������, 2025

    2025

  62. [70]

    R1-vl: Learning to reason with multimodal large language models via step-wise group relative policy optimization

    Jingyi Zhang, Jiaxing Huang, Huanjin Yao, Shunyu Liu, Xikun Zhang, Shijian Lu, and Dacheng Tao. R1-vl: Learning to reason with multimodal large language models via step-wise group relative policy optimization. ����� �������� ����������������, 2025

    2025

  63. [71]

    Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? In �������� ���������� �� �������� ������, pages 169–186

    Renrui Zhang, Dongzhi Jiang, Yichi Zhang, Haokun Lin, Ziyu Guo, Pengshuo Qiu, Aojun Zhou, Pan Lu, Kai-Wei Chang, Yu Qiao, et al. Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? In �������� ���������� �� �������� ������, pages 169–186. Sprin...

    2024

  64. [72]

    Mavis: Mathematical visual instruction tuning with an automatic data engine

    Renrui Zhang, Xinyu Wei, Dongzhi Jiang, Ziyu Guo, Shicheng Li, Yichi Zhang, Chengzhuo Tong, Jiaming Liu, Aojun Zhou, Bin Wei, et al. Mavis: Mathematical visual instruction tuning with an automatic data engine. ����� �������� ����������������, 2024

    2024

  65. [73]

    thinking with images

    Ziwei Zheng, Michael Yang, Jack Hong, Chenxiao Zhao, Guohai Xu, Le Yang, Chao Shen, and Xing Yu. Deepeyes: Incentivizing" thinking with images" via reinforcement learning. ����� �������� ����������������, 2025. 15

    2025

  66. [74]

    Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models

    Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shenglong Ye, Lixin Gu, Yuchen Duan, Hao Tian, Weijie Su, Jie Shao, et al. Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models. ����� �������� ����������������, 2025

    2025

  67. [75]

    Math-puma: Progressive upward multimodal alignment to enhance mathematical reasoning

    Wenwen Zhuang, Xin Huang, Xiantao Zhang, and Jin Zeng. Math-puma: Progressive upward multimodal alignment to enhance mathematical reasoning. ����� �������� ����������������, 2024

    2024

  68. [76]

    Math-puma: Progressive up- ward multimodal alignment to enhance mathematical reasoning

    Wenwen Zhuang, Xin Huang, Xiantao Zhang, and Jin Zeng. Math-puma: Progressive up- ward multimodal alignment to enhance mathematical reasoning. In ����������� �� ��� ���� ���������� �� ��������� ������������, volume 39, pages 26183–26191, 2025

    2025

  69. [77]

    multi-images per question

    Chengke Zou, Xingang Guo, Rui Yang, Junyu Zhang, Bin Hu, and Huan Zhang. Dynamath: A dynamic visual benchmark for evaluating mathematical reasoning robustness of vision language models. ����� �������� ����������������, 2024. 16 Contents 1 Introduction 1 2 Related Work 3 3 W E-...

    2024

Pith tools

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