Pith. sign in

REVIEW 3 major objections 6 minor 3 cited by

MiCo: Multi-image Contrast for Reinforcement Visual Reasoning

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read MiCo shows that a 7B vision-language model trained only on same/different image triplets with rule-based reinforcement learning outperforms GPT-4o on VLM2-Bench without any human-annotated QA pairs.

desk verdict Genuinely new idea for annotation-free multi-image reasoning supervision, with large reported gains, but the Augmented GRPO objective as written is not a valid estimator and the empirical claims rest on an unreleased implementation. read the letter →

arxiv 2506.22434 v1 pith:TQMR47DM submitted 2025-06-27 cs.CV

Xi Chen , Mingkang Zhu , Shaoteng Liu , Xiaoyang Wu , Xiaogang Xu , Yu Liu , Xiang Bai , Hengshuang Zhao This is my paper
classification cs.CV
keywords multi-imagereasoningvision-languagemodelsreinforcementlearningchain-of-thoughtself-supervisedvisualcomparisonrule-basedRLdataaugmentation
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

This paper tries to show that the main thing vision-language models lack for multi-image understanding is a general visual-comparison skill, and that this skill can be trained from images alone, without human-written question-answer pairs. The method builds training triplets from two augmented views of the same image plus a similar but distinct third image, asks the model to reason about which views are the same, and optimizes with rule-based reinforcement learning under a weak-to-strong augmentation scheme. The reported result is that this comparison-only training transfers to a wide range of multi-image and general vision benchmarks, including a new state of the art on VLM2-Bench with a 7B model that surpasses GPT-4o. If correct, it offers an annotation-free route to multi-image reasoning and sharpens the claim that visual comparison is the core bottleneck.

What carries the argument

The central object is the contrastive triplet, a triple of images $T = \{T_1(I_a), T_2(I_a), T_3(I_b)\}$ in which the first two are different random crops of the same image and the third is a visually similar but distinct image, each augmented by cropping and resizing. The mechanism that carries the argument is Augmented GRPO: chain-of-thought rollouts are sampled under weak augmentation, reward is computed by a rule-based check of format and of the correct same/different labels for all queried pairs, and the policy is updated under a stronger augmentation of the same triplets. This weak-to-strong optimization forces the model to produce reliable comparisons that survive harder visual transforms, and the contrastive structure discourages shortcut learning by making the discrimination depend on fine detail. Data sources are filtered so negatives are subtle: video frames separated by about two seconds with SSIM-based removal of near-identical pairs, and image-editing before/after pairs with pixel-level filtering of overly large differences.

What would settle it

Take MiCo's strongly augmented training triplets and have human annotators judge same/different on each; if a substantial fraction (say, more than 10%) are judged visually indeterminate, then a portion of the reward signal is wrong, and training should degrade as that fraction grows. A cleaner test would be to train two copies of MiCo, one with strong augmentations that always crop inside the discriminative region and one that never does, and compare their VLM2-Bench scores.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that visual comparison can be treated as a self-supervised meta-skill: by constructing contrastive triplets from video frames and image-editing pairs, and by using rule-based reinforcement learning with group-relative advantages, a vision-language model can be incentivized to produce chain-of-thought comparisons and attend to subtle cross-image differences. The paper reports an average of 61.06 on VLM2-Bench for MiCo-7B-CoT, an improvement of +12.93 over the Qwen2.5-VL-7B baseline, surpassing GPT-4o's 60.36. Generalization is documented on MuirBench, BLINK, MMStar, MMMU, HallusionBench, and MathVista, with the largest gains on correspondence-style tasks such as visual retrieval and semantic correspondence. The paper attributes the transfer to the learned comparison ability rather than to task-specific supervision, and it notes that face verification, visual math, and relative depth remain weak points.

Load-bearing premise

Random cropping and resizing are assumed to preserve the subtle difference that determines whether two images are the same, so the rule-based reward labels remain correct even after strong augmentation; if a strong crop removes the only distinguishing detail, the model is rewarded for an impossible discrimination.

Editorial extensions

If this is right

  • A 7B vision-language model trained only on comparison triplets can surpass closed-source models on VLM2-Bench, showing that annotation-free rule-based RL scales multi-image reasoning.
  • The learned comparison skill transfers without retraining to tasks such as object matching, tracking, semantic correspondence, spatial relations, and some single-image reasoning, which the paper documents across six benchmarks.
  • Combining two heterogeneous contrastive data sources (video frames and image-editing pairs) is better than either alone, and the method generalizes across different editing datasets.
  • Tasks that resist verbal description, such as face identity verification, relative depth, and visual math, are not fixed by this training, marking the boundary of the claimed skill.

Reading between the lines

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

  • If visual comparison is truly the bottleneck skill, then scaling triplet difficulty and diversity—harder negatives, longer temporal gaps, more editing operations—should keep lifting multi-image benchmarks, which the paper's data-source ablation suggests.
  • The weak-to-strong rollout trick is a general curriculum recipe: sample reasoning traces where the reward is easy to achieve, then reinforce them under harder conditions; it could be applied to other self-supervised reinforcement learning problems beyond same/different comparison.
  • The face-verification failure suggests that language-based chain-of-thought may not carry purely perceptual identity information; a parallel non-verbal objective, such as a learned similarity head, might complement chain-of-thought rather than replace it.
  • A direct stress test of the core assumption would be to replace random cropping with region-dropping augmentations and measure whether the reward remains correct; the paper does not report this ablation.
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 / 6 minor

Summary. The paper proposes MiCo, a self-supervised reinforcement-learning framework for multi-image reasoning in VLMs. Training triplets are built from two augmented views of the same image plus a visually similar but distinct image, sourced from video frames and image editing datasets. The VLM is prompted to output a chain of thought and answer same/different questions, with rule-based rewards. The authors introduce "Augmented GRPO," in which rollouts are sampled under weak augmentations and the policy is optimized on strong augmentations. Experiments on VLM2-Bench report a +12.93 average improvement over Qwen2.5-VL-7B, surpassing GPT-4o, and additional gains on several multi-image and single-image benchmarks.

Significance. The work addresses an important problem: improving cross-image reasoning without human-annotated question-answer pairs. The central idea of using inherent image constraints as reward supervision is attractive and potentially scalable. The paper also merits credit for systematic ablations (Table 2) and for evaluating on a range of benchmarks, including task-level analyses. If the method is valid, MiCo would be a promising annotation-free route to multi-image reasoning. However, the central algorithmic contribution, Augmented GRPO, is described with a mathematically invalid objective, which undermines the claimed mechanism and the interpretation of the key ablation. The paper needs a major revision to clarify or correct the training objective before its empirical claims can be fully credited.

major comments (3)
  1. [Algorithm 1, Eq. (1), Section 3.3] The Augmented GRPO objective as written is not a valid policy-gradient estimator. In Algorithm 1, rollouts o_i are sampled from π_θold(·|q_w) and rewarded on q_w, but the loss in line 11 uses the importance ratio π_θ(o_i|q_s)/π_θold(o_i|q_s) for the strong prompt q_s. Standard PPO/GRPO requires the rollouts to be sampled from the same distribution that appears in the ratio. Without a correction such as π_θ(o|q_s)/π_θold(o|q_w), the expectation is over q_w while the ratio is for q_s, so the gradient estimator is not justified. The importance-sampling attempt in Appendix B.2 reweights advantages instead of correcting the sampling distribution and is reported as unhelpful, which does not repair the derivation. Consequently, the central claim that the policy is "optimized on stronger augmented contexts" is not supported by the written objective, and Table 2(c) cannot separate the intended curriculum from this bias. Since code and data are not released, the empirical results cannot confirm which objective was actually optimized. Please either provide a correct derivation (e.g., a valid off-policy correction or a properly formulated weak-to-strong objective) or explicitly present the loss as a heuristic weighted regression and provide empirical justification for that heuristic without invoking policy-gradient estimation.
  2. [Section 3.2, Figure 2, Algorithm 1 lines 4-5] The method assumes that random cropping and resizing preserve the subtle visual differences that determine same/different, so that the rule-based reward is correct under strong augmentation. This assumption is load-bearing but untested. If a strong crop removes the only discriminative region, the reward labels the model's answer as incorrect even though the images are visually indistinguishable, corrupting the training signal. The ablations in Table 2(f) compare augmentation types on benchmark performance but do not measure label-preservation rates or reward noise under strong augmentation. Please include a diagnostic such as human agreement, a vision model's comparison accuracy, or a quantitative measure of discriminative-detail preservation under the chosen augmentation distributions.
  3. [Table 2(c), Section 4.3] Even if the objective were corrected, the comparison (Weak, Strong) versus (Strong, Strong) versus (Weak, Weak) confounds rollout quality, reward noise, and optimization difficulty. Strong augmentation may yield many zero-reward rollouts, while weak augmentation yields higher-quality but potentially less challenging trajectories. The current experiment does not isolate the curriculum effect from these confounds. An additional control, such as using weak rollouts with a properly reweighted objective on weak prompts, or filtering/re-labeling strong rollouts, would be needed to support the claim that the gains arise from the proposed weak-to-strong transfer.
minor comments (6)
  1. [Section 3.3] Typo: "reasnoning prompt" should be "reasoning prompt".
  2. [Eq. (1)] The displayed objective is malformed: the expectation is written without an integrand, and the notation mixes q with q_s/q_w. Please rewrite the equation to match the algorithm.
  3. [Appendix A] Typo: "image edting data" should be "image editing data".
  4. [Section 4.5] Typo: "successfuly" should be "successfully".
  5. [Algorithm 1] The notation T^w(I_1,I_2,I_3) is defined earlier as T = {T_1(I_a), T_2(I_a), T_3(I_b)}; please clarify the relationship between the triplet notation and the per-image augmentation operators.
  6. [Section 3.3] The paper says GPT-4o is used to expand user questions but does not provide the expanded prompt set; please release the prompt templates or include them in an appendix for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: MiCo's self-supervised contrastive rewards are independent of its external evaluation benchmarks.

full rationale

MiCo's training signal is generated from image triplets: two augmented views of the same image and a third, visually similar but distinct image, with the same/different ground truth determined by image source rather than by any benchmark answer. The reward is rule-based (format reward plus correctness of the T/F comparisons), and no parameter is fitted to VLM2-Bench, MuirBench, BLINK, or the other evaluation suites. The claim that visual comparison training generalizes to other multi-image and single-image tasks is therefore supported by out-of-distribution evaluation, not by construction. The paper cites prior work on GRPO, contrastive learning, and reasoning VLMs, but none of these citations is a self-citation that carries the central load; there is no imported uniqueness theorem and no fitted quantity that is renamed as a prediction. The Augmented GRPO rollout/augmentation mismatch noted by the skeptic is a methodological or correctness concern about the policy-gradient estimator, not a circularity of the kind where the derivation reduces to its own inputs. Accordingly, there is no significant circularity; score 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method relies on hand-chosen data filtering thresholds and augmentation choices, plus several domain assumptions about what drives multi-image reasoning and what the augmentations preserve. The most consequential assumption, augmentation-invariant labels, is not explicitly tested.

free parameters (5)
  • pixel_difference_ratio_threshold = 0.8
    Hand-chosen threshold in Appendix A; pairs with >0.8 differing pixels are discarded. Not grounded in theory.
  • ssim_similarity_threshold = 0.95
    Hand-chosen SSIM cutoff in Appendix A to remove near-identical video frame pairs.
  • temporal_gap = 2 seconds
    Video frame pairs sampled with 2s gap (Section 3.2); controls difficulty of negative pairs.
  • augmentation_prompt_variation_count = 50 (per Table 2e)
    Number of GPT-4o expanded prompt templates; chosen empirically, with saturation around 50.
  • crop_resize_parameters = unspecified
    Random cropping and resizing is the default augmentation (Section 3.2, Table 2f), but the exact scale/range is not stated.
assumptions (4)
  • domain assumption Visual comparison is the core meta-skill required for multi-image reasoning
    Section 3.1-3.2: The method targets comparison as the key deficit; if multi-image failure is dominated by other factors, the approach underdelivers.
  • domain assumption Random cropping and resizing preserves the discriminative information needed to answer same/different correctly
    Section 3.2 and Algorithm 1: The ground-truth reward assumes that weak and strong augmentations do not change the correct answer; a crop that removes the only distinguishing detail corrupts the label.
  • domain assumption The rule-based reward (format + accuracy) is a sufficient training signal for CoT reasoning
    Section 3.3: Borrowed from DeepSeek-R1; the paper provides no analysis of reward noise or whether binary T/F rewards give dense enough signal.
  • domain assumption Qwen2.5-VL-7B already has strong single-image perception, so the remaining deficit is cross-image comparison
    Section 3.1: Motivates the training task; if the base model's single-image perception is also weak, the comparison-only training may not transfer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MiCo: Multi-image Contrast for Reinforcement Visual Reasoning." pith.science (2026). https://pith.science/paper/TQMR47DM

@misc{pith2026250622434,
  author       = {Pith},
  title        = {Pith review of: MiCo: Multi-image Contrast for Reinforcement Visual Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TQMR47DM}},
  note         = {Machine review of arXiv:2506.22434}
}
read the original abstract

This work explores enabling Chain-of-Thought (CoT) reasoning to link visual cues across multiple images. A straightforward solution is to adapt rule-based reinforcement learning for Vision-Language Models (VLMs). However, such methods typically rely on manually curated question-answer pairs, which can be particularly challenging when dealing with fine grained visual details and complex logic across images. Inspired by self-supervised visual representation learning, we observe that images contain inherent constraints that can serve as supervision. Based on this insight, we construct image triplets comprising two augmented views of the same image and a third, similar but distinct image. During training, the model is prompted to generate a reasoning process to compare these images (i.e., determine same or different). Then we optimize the model with rule-based reinforcement learning. Due to the high visual similarity and the presence of augmentations, the model must attend to subtle visual changes and perform logical reasoning to succeed. Experiments show that, although trained solely on visual comparison tasks, the learned reasoning ability generalizes effectively to a wide range of questions. Without relying on any human-annotated question-answer pairs, our method achieves significant improvements on multi-image reasoning benchmarks and shows strong performance on general vision tasks.

Figures

Figures reproduced from arXiv: 2506.22434 by the authors.

Figure 1
Figure 1. Challenges for multi-image understanding. While recent works support multiple images as input, most of them focus on scenarios where each image can be interpreted independently (e.g., Example 1), which remains relatively easy for current state-of-the-art VLMs. However, many real￾world tasks (e.g., Example 2-4) require models to compare subtle visual differences, align visual cues across images, and reason about obje… view at source ↗
Figure 2
Figure 2. Demonstrations for contrastive samples. The first row shows two triplets from the video, and the second row demonstrates samples from image editing datasets. These samples are visually similar but contain subtle differences (marked with red circles), on which we apply random cropping and resizing. In each triplet, the first two images are the same, and the third image is different. Image selection. We begin by selec… view at source ↗
Figure 3
Figure 3. Demonstrations for visual reasoning. Given a question, MiCo first examines the details of each image to identify answer-related visual cues, and then performs cross-image comparisons to derive the final answer. The reasoning processes are marked in gray, with key contents underlined. 4.5 Qualitative Analysis We provide qualitative demonstrations in [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Demonstrations for detailed comparison and jigsaw solving. exploited in privacy-invading scenarios if deployed irresponsibly. To mitigate such risks, we advocate for deploying MiCo in alignment with responsible AI guidelines, ensuring transparency, consent, and clear b…
Figure 5
Figure 5. Figure 5: Demonstrations for IQ test, functional correspondence, and visual similarity. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

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

  1. MARS-RA: Rank Aggregation for Credit Assignment via Multimodal Comparisons in Embodied Multi-Agent Cooperation

    cs.AI 2026-07 conditional novelty 6.0 of 10

    Credit assignment via LMM pairwise comparisons plus Bradley–Terry rank aggregation and potential-based shaping improves cooperative MARL under sparse rewards and dynamic agent counts.

  2. DataClaw0: Agentic Tailoring Multimodal Data from Raw Streams

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    DataClaw0 introduces an agentic data-tailoring paradigm, a 9B model trained on a synthetically generated dataset, and a new benchmark, claiming improved downstream adaptation in video generation, VQA, and GUI navigati...

  3. ReMoT: Reinforcement Learning with Motion Contrast Triplets

    cs.CV 2026-02 conditional novelty 6.0 of 10

    Training a 4B vision-language model on rule-generated motion-contrast triplets with GRPO lifts spatio-temporal QA accuracy by about 17 points on the authors' own benchmark and by smaller margins on standard benchmarks.

Reference graph

Works this paper leans on

46 extracted references · 23 canonical work pages · cited by 3 Pith papers

  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. arXiv:2303.08774, 2023. 1, 2

    arXiv 2023

  2. [2]

    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.arXiv:2502.13923, 2025. 1, 2, 5, 6, 7, 8

    arXiv 2025

  3. [3]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herve Jegou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InICCV, 2021. 2

    work page 2021

  4. [4]

    Sft or rl? an early investigation into training r1-like reasoning large vision-language models.arXiv:2504.11468,

    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.arXiv:2504.11468,

    arXiv

  5. [5]

    Liang Chen, Lei Li, Haozhe Zhao, and Yifan Song. Vinci. r1-v: Reinforcing super generalization ability in vision-language models with less than $3.https://github.com/Deep-Agent/R1-V, 2025. 1, 2

    work page 2025

  6. [6]

    Are we on the right way for evaluating large vision-language models?NeurIPS,

    Lin Chen, Jinsong Li, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Zehui Chen, Haodong Duan, Jiaqi Wang, Yu Qiao, Dahua Lin, et al. Are we on the right way for evaluating large vision-language models?NeurIPS,

    work page

  7. [7]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. InICML, 2020. 2

    work page 2020

  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, et al. Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks. InCVPR, 2024. 1, 3, 6

    work page 2024

Show all 46 references
  1. [9]

    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. InACMMM, 2024. 6

    2024

  2. [10]

    Blink: Multimodal large language models can see but not perceive

    Xingyu Fu, Yushi Hu, Bangzheng Li, Yu Feng, Haoyu Wang, Xudong Lin, Dan Roth, Noah A Smith, Wei-Chiu Ma, and Ranjay Krishna. Blink: Multimodal large language models can see but not perceive. In ECCV, 2024. 2, 8, 14

    2024

  3. [11]

    Hallusionbench: an advanced diagnostic suite for entangled language hallucination and visual illusion in large vision-language models

    Tianrui Guan, Fuxiao Liu, Xiyang Wu, Ruiqi Xian, Zongxia Li, Xiaoyu Liu, Xijun Wang, Lichang Chen, Furong Huang, Yaser Yacoob, et al. Hallusionbench: an advanced diagnostic suite for entangled language hallucination and visual illusion in large vision-language models. InCVPR, ...

    2024

  4. [12]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv:2501.12948, 2025

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv:2501.12948, 2025. 1, 2, 5

    2025 arXiv

  5. [13]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollar, and Ross Girshick. Masked autoencoders are scalable vision learners. InCVPR, 2022. 2

    2022

  6. [14]

    Momentum contrast for unsupervised visual representation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual representation learning. InCVPR, 2020. 2

    2020

  7. [15]

    Gpt-4o system card.arXiv:2410.21276, 2024

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv:2410.21276, 2024. 1, 2, 3, 4, 6

    2024 arXiv

  8. [16]

    Openai o1 system card.arXiv:2412.16720, 2024

    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.arXiv:2412.16720, 2024. 1, 2

    2024 arXiv

  9. [17]

    Llava-onevision: Easy visual task transfer.TMLR, 2025

    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.TMLR, 2025. 2, 6

    2025

  10. [18]

    Llava- next-interleave: Tackling multi-image, video, and 3d in large multimodal models.arXiv:2407.07895, 2024

    Feng Li, Renrui Zhang, Hao Zhang, Yuanhan Zhang, Bo Li, Wei Li, Zejun Ma, and Chunyuan Li. Llava- next-interleave: Tackling multi-image, video, and 3d in large multimodal models.arXiv:2407.07895, 2024. 2

    2024 arXiv

  11. [19]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. InNeurIPS, 2023. 1, 2 10

    2023

  12. [20]

    Noisyrollout: Reinforcing visual reasoning with data augmentation.arXiv:2504.13055, 2025

    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.arXiv:2504.13055, 2025. 2, 5, 6, 8, 13

    2025

  13. [21]

    Mmdu: A multi-turn multi-image dialog understanding benchmark and instruction-tuning dataset for lvlms

    Ziyu Liu, Tao Chu, Yuhang Zang, Xilin Wei, Xiaoyi Dong, Pan Zhang, Zijian Liang, Yuanjun Xiong, Yu Qiao, Dahua Lin, et al. Mmdu: A multi-turn multi-image dialog understanding benchmark and instruction-tuning dataset for lvlms. InNeurIPS, 2025. 2

    2025

  14. [22]

    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. InICLR, 2024. 8

    2024

  15. [23]

    Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforcement learning.arXiv:2503.07365, 2025

    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.arXiv:2503.07365, 2025. 1, 2, 5, 6, 8

    2025 arXiv

  16. [24]

    Mmiu: Multimodal multi-image understanding for evaluating large vision- language models

    Fanqing Meng, Jin Wang, Chuanhao Li, Quanfeng Lu, Hao Tian, Jiaqi Liao, Xizhou Zhu, Jifeng Dai, Yu Qiao, Ping Luo, et al. Mmiu: Multimodal multi-image understanding for evaluating large vision- language models. InICLR, 2025. 2, 3

    2025

  17. [25]

    Skywork r1v: Pioneering multimodal reasoning with chain-of-thought

    Yi Peng, Xiaokun Wang, Yichen Wei, Jiangbo Pei, Weijie Qiu, Ai Jian, Yunzhuo Hao, Jiachun Pan, Tianyidan Xie, Li Ge, et al. Skywork r1v: Pioneering multimodal reasoning with chain-of-thought. arXiv:2504.05599, 2025. 2

    2025 arXiv

  18. [26]

    Lmm-r1: Empowering 3b lmms with strong reasoning abilities through two-stage rule-based rl.arXiv:2503.07536, 2025

    Yingzhe Peng, Gongrui Zhang, Miaosen Zhang, Zhiyuan You, Jie Liu, Qipeng Zhu, Kai Yang, Xingzhong Xu, Xin Geng, and Xu Yang. Lmm-r1: Empowering 3b lmms with strong reasoning abilities through two-stage rule-based rl.arXiv:2503.07536, 2025. 2

    2025 arXiv

  19. [27]

    Seed-thinking-v1

    ByteDance Seed, Yufeng Yuan, Yu Yue, Mingxuan Wang, Xiaochen Zuo, Jiaze Chen, Lin Yan, Wenyuan Xu, Chi Zhang, Xin Liu, et al. Seed-thinking-v1. 5: Advancing superb reasoning models with reinforcement learning.arXiv:2504.13914, 2025. 1, 2

    2025

  20. [28]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv:2402.03300, 2024

    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.arXiv:2402.03300, 2024. 1, 5, 7, 13

    2024 arXiv

  21. [29]

    Reason-rft: Reinforcement fine-tuning for visual reasoning.arXiv:2503.20752, 2025

    Huajie Tan, Yuheng Ji, Xiaoshuai Hao, Minglan Lin, Pengwei Wang, Zhongyuan Wang, and Shanghang Zhang. Reason-rft: Reinforcement fine-tuning for visual reasoning.arXiv:2503.20752, 2025. 2

    2025

  22. [30]

    Vidgen-1m: A large-scale dataset for text-to-video generation.arXiv:2408.02629, 2024

    Zhiyu Tan, Xiaomeng Yang, Luozheng Qin, and Hao Li. Vidgen-1m: A large-scale dataset for text-to-video generation.arXiv:2408.02629, 2024. 5, 7

    2024 arXiv

  23. [31]

    Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, et al. Kimi k1. 5: Scaling reinforcement learning with llms. arXiv:2501.12599, 2025. 1, 2

    2025 arXiv

  24. [32]

    Muirbench: A comprehensive benchmark for robust multi-image understanding

    Fei Wang, Xingyu Fu, James Y Huang, Zekun Li, Qin Liu, Xiaogeng Liu, Mingyu Derek Ma, Nan Xu, Wenxuan Zhou, Kai Zhang, et al. Muirbench: A comprehensive benchmark for robust multi-image understanding. InICLR, 2025. 2, 3, 8, 14

    2025

  25. [33]

    Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution.arXiv:2409.12191, 2024

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, et al. Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution.arXiv:2409.12191, 2024. 3, 6

    2024 arXiv

  26. [34]

    Sota with less: Mcts-guided sample selection for data-efficient visual reasoning self-improvement.arXiv:2504.07934, 2025

    Xiyao Wang, Zhengyuan Yang, Chao Feng, Hongjin Lu, Linjie Li, Chung-Ching Lin, Kevin Lin, Furong Huang, and Lijuan Wang. Sota with less: Mcts-guided sample selection for data-efficient visual reasoning self-improvement.arXiv:2504.07934, 2025. 1, 2, 5, 6, 8

    2025 arXiv

  27. [35]

    Omniedit: Building image editing generalist models through specialist supervision

    Cong Wei, Zheyang Xiong, Weiming Ren, Xeron Du, Ge Zhang, and Wenhu Chen. Omniedit: Building image editing generalist models through specialist supervision. InICLR, 2024. 5, 7

    2024

  28. [36]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. InNeurIPS, 2022. 2

    2022

  29. [37]

    Towards open-ended visual quality comparison

    Haoning Wu, Hanwei Zhu, Zicheng Zhang, Erli Zhang, Chaofeng Chen, Liang Liao, Chunyi Li, Annan Wang, Wenxiu Sun, Qiong Yan, et al. Towards open-ended visual quality comparison. InECCV, 2024. 2

    2024

  30. [38]

    mplug-owl2: Revolutionizing multi-modal large language model with modality collaboration

    Qinghao Ye, Haiyang Xu, Jiabo Ye, Ming Yan, Anwen Hu, Haowei Liu, Qi Qian, Ji Zhang, and Fei Huang. mplug-owl2: Revolutionizing multi-modal large language model with modality collaboration. InCVPR,

  31. [39]

    Mmmu: A massive multi-discipline multimodal understanding and reasoning benchmark for expert agi

    Xiang Yue, Yuansheng Ni, Kai Zhang, Tianyu Zheng, Ruoqi Liu, Ge Zhang, Samuel Stevens, Dongfu Jiang, Weiming Ren, Yuxuan Sun, et al. Mmmu: A massive multi-discipline multimodal understanding and reasoning benchmark for expert agi. InCVPR, 2024. 2, 8

    2024

  32. [40]

    Vlm2-bench: A closer look at how well vlms implicitly link explicit matching visual cues.arXiv:2502.12084, 2025

    Jianshu Zhang, Dongyu Yao, Renjie Pi, Paul Pu Liang, and Yi R Fung. Vlm2-bench: A closer look at how well vlms implicitly link explicit matching visual cues.arXiv:2502.12084, 2025. 2, 3, 6, 7

    2025 arXiv

  33. [41]

    Weaving context across images: Improving vision-language models through focus-centric visual chains.arXiv:2504.20199, 2025

    Juntian Zhang, Yuhan Liu, Wei Liu, Jian Luan, Rui Yan, et al. Weaving context across images: Improving vision-language models through focus-centric visual chains.arXiv:2504.20199, 2025. 2

    2025 arXiv

  34. [42]

    Long context transfer from language to vision.arXiv:2406.16852,

    Peiyuan Zhang, Kaichen Zhang, Bo Li, Guangtao Zeng, Jingkang Yang, Yuanhan Zhang, Ziyue Wang, Haoran Tan, Chunyuan Li, and Ziwei Liu. Long context transfer from language to vision.arXiv:2406.16852,

    arXiv

  35. [43]

    Video instruction tuning with synthetic data.arXiv:2410.02713, 2024

    Yuanhan Zhang, Jinming Wu, Wei Li, Bo Li, Zejun Ma, Ziwei Liu, and Chunyuan Li. Video instruction tuning with synthetic data.arXiv:2410.02713, 2024. 6

    2024 arXiv

  36. [44]

    Benchmarking multi-image understanding in vision and language models: Perception, knowledge, reasoning, and multi-hop reasoning

    Bingchen Zhao, Yongshuo Zong, Letian Zhang, and Timothy Hospedales. Benchmarking multi-image understanding in vision and language models: Perception, knowledge, reasoning, and multi-hop reasoning. arXiv:2406.12742, 2024. 2, 3

    2024 arXiv

  37. [45]

    Ultraedit: Instruction-based fine-grained image editing at scale

    Haozhe Zhao, Xiaojian Shawn Ma, Liang Chen, Shuzheng Si, Rujie Wu, Kaikai An, Peiyu Yu, Minjia Zhang, Qing Li, and Baobao Chang. Ultraedit: Instruction-based fine-grained image editing at scale. In NeurIPS, 2024. 7

    2024

  38. [46]

    Are image1 and image2 the same?

    Yikang Zhou, Tao Zhang, Shilin Xu, Shihao Chen, Qianyu Zhou, Yunhai Tong, Shunping Ji, Jiangning Zhang, Xiangtai Li, and Lu Qi. Are they the same? exploring visual correspondence shortcomings of multimodal llms.arXiv:2501.04670, 2025. 2 12 A More Implementation Details Prompt ...

    2025 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.