betaFeedback

WorksComputer Science

ViRA: Toward Ultra-High-Resolution Visual Agents via Cross-Input Relative Policy Optimization

Abstract

Visual agents answer questions about ultra-high-resolution photos by zooming into the relevant regions. However, existing agents often answer from common sense instead of reading the evidence. Reinforcement learning (RL) on the final answer cannot correct this, because a guess earns the same reward as reading. We propose Cross-Input Relative Policy Optimization (CrossRPO), which trains agents on contrast sets. A contrast set asks one question about several versions of an image that differ only in a small region with the evidence and have different answers, so only reading the evidence solves the whole set. CrossRPO rewards solving the whole set, and we prove that it does not reinforce a set answered from common sense. We build contrast sets for training (ViRA-Train) and evaluation (ViRA-Bench) by editing one small object in images of 21 to 120 megapixels, and we photograph further sets with real changes (ViRA-Bench-Real) to exclude editing traces. We train two Visual Reasoning Agents, ViRA-9B and ViRA-27B, with CrossRPO on only about 3,500 items of ViRA-Train each. CrossRPO achieves 63.3% set accuracy on ViRA-Bench at the maximum resolution, against 59.3% for GRPO. ViRA-9B is the most accurate open model of its size on all benchmarks, and ViRA-27B outperforms Claude Haiku 5.5 and GPT-6-Luna, both at their highest thinking level, on most benchmarks.

CODE link text

DATA link text

MODELS link text link text

BENCH link text

Introduction

Photos from modern cameras have tens of megapixels (MP), but the detail that answers a question about them is often only a few dozen pixels wide, such as what a pedestrian down the street is holding. Such photos exceed the input size of vision-language models. For example, Qwen3.5 turns each patch of 32×3232 \times32 pixels into one token, so a 45 MP photo would require about 44,000 tokens. Models therefore cap the input size, even with high-resolution encoders [3, 22, 39, 61], and shrink larger photos to fit. The detail shrinks with the photo and often becomes unreadable [66, 73, 84]. At the 3.8 MP limit of Claude Haiku 5.5, a detail 50 pixels wide in a 45 MP photo becomes about 15 pixels wide. To recover such details, visual agents view the whole photo at low resolution and zoom into the regions they need [50, 60]. Some follow a fixed search procedure [35, 40, 41, 57, 67, 82], and others learn where to zoom [13, 25, 33, 53, 59, 83, 85, 87].

Correct without reading. However, a correct answer does not show that the agent read the evidence, because common sense often gives the same answer [1, 19, 38, 49, 64, 75]. For example, a person who holds up a small device as if taking a photo usually holds a smartphone. An agent can therefore answer “smartphone” correctly without reading the device. Figure 1a tests this with three versions of a 45 MP street photo, in which a woman holds up a smartphone, a calculator or a TV remote. The device fits in a square of 110 pixels, so a model must zoom in to read it. Without a zoom tool, Claude Haiku 5.5 and GPT-6-Luna answer “smartphone” for all three versions. The released agents DeepEyes [87] and ZoomEye [57] give the same answer, although they can zoom. Even with a zoom tool and its highest thinking level, Claude Haiku 5.5 still identifies the remote as a smartphone. Being able to zoom therefore does not make an agent read the evidence.

Figure 1. ViRA-27B leads at every budget. (a) Three versions of a 45-megapixel (MP) ViRA-Bench photo differ only in the device a woman holds up (yellow box). Rows give models’ answers at 1 MP budget, and Set is correct only if all three are. (b) Set accuracy is the share of sets with all versions correct, and tokens are summed over all turns and versions. Filled markers are pixel budgets of 0.5–4 MP, which cap every image a model sees, and hollow markers are models’ max resolution. Claude Haiku 5.5 has no 4 MP marker as its max resolution is 3.8 MP. Max thinking is the highest thinking level, and with zoom tool, a model draws a box to see that region.

Existing methods. The agents that learn where to zoom are trained with reinforcement learning (RL) on the final answer [21, 29], typically with Group Relative Policy Optimization (GRPO) [55]. This training cannot correct the guess, because on a single image, a guess earns the same reward as reading. The guess is also easier, because it requires no zoom, so it becomes a shortcut that the reward cannot detect [10, 16, 77, 79]. Recent work therefore also rewards the zoom, such as a tool call [87], a crop that covers annotated evidence [11, 86] or a call that improves the answer [14, 51]. These rewards credit viewing the evidence rather than reading it, and agents learn to call the zoom tool without using what it shows [25, 47, 69, 90]. For example, on the TV remote version of Figure 1a, DeepEyes zooms onto the whole remote but writes that “the zoomed-in image confirms that the person is holding a smartphone.” Given only this zoom in a new conversation, it answers “TV remote.” Other methods mask the evidence and reward abstention or penalize an answer [42, 74]. However, a masked image has no answer to read, so they cannot check whether the answer changes with the evidence. Both still judge one image at a time.

Contrast sets. On one image, reading and guessing give the same answer, so no reward on that image can separate them. Several versions of the same photo can. A contrast set [15] asks the same question about several versions of a photo. Only a small region with the evidence differs between them, so the versions have different answers, as in Figure 1a. An agent that gives the same answer to every version can be correct on only one of them, so only reading the evidence can solve the whole set. We call the share of sets that an agent solves its set accuracy. Contrast sets and similar minimal pairs have been used to evaluate vision-language models [17, 34, 63] and to train them with supervised losses [7, 31, 62]. We build such sets for zoom agents by editing one small object in real photos of 21 to 120 MP. ViRA-Train has 9,261 sets with K=1K = 1, 2 or 3 versions for training, and ViRA-Bench has 762 sets for evaluation. In ViRA-Bench, the edited object covers a median of 0.0075% of the photo. We also photograph 78 sets in which only the real object changes between shots (ViRA-Bench-Real), so they involve no editing. Figure 2 shows examples of both benchmarks, on which many of the models we test give the same answer to all three versions.

Figure 2. Examples of ViRA-Bench and ViRA-Bench-Real. Each set shows the original photo (O, cut to 16:9) with the object in a yellow box. The insets show the object in O and the two variants (A, B), with the correct answers. Below each set are the original’s answer and the number of baseline settings, out of 15 (Appendix § 11.§ 1), that give this answer to all three versions at 1 MP. Questions are shortened.

CrossRPO. GRPO samples several answers (rollouts) for one input and compares them with each other (Figure 3, top). If it trains on each version of a contrast set separately, no reward depends on solving the whole set. If it compares rollouts across versions (pooled GRPO), a guess that is correct on one version scores above the wrong answers on the others and is rewarded. We therefore propose Cross-Input Relative Policy Optimization (CrossRPO, Figure 3, bottom). It compares each rollout only with rollouts of the same version, so a guess that is always correct on one version receives no credit. It then scales the result with a cross-input weight. The weight grows when the other versions are answered correctly, because a correct answer on this version then solves the set. CrossRPO applies to sets of any size, including single images, whereas recent methods for two inputs [12, 27, 37, 89, 91] handle only pairs. We prove that its update is unbiased and does not reinforce a set answered from common sense, and that the objectives of GRPO and of these two-input methods are special cases of its objective.

Figure 3. GRPO and CrossRPO. The policy πθ\pi_\theta samples the rollouts, and the reward rr is the correctness cc of the answer. (a) GRPO compares the rollouts of one input ss. (b) CrossRPO compares rollouts within each version svs_v of a set and multiplies the result by 1+λwv1+\lambda w_v, where wvw_v comes from the rewards of the other versions.

ViRA. With CrossRPO, we train two Visual Reasoning Agents, ViRA-9B and ViRA-27B, on about 3,500 items of ViRA-Train each. With the same data and reward, CrossRPO achieves 63.3% set accuracy on ViRA-Bench at the maximum resolution, against 59.3% for GRPO on each version separately and 52.6% for pooled GRPO. Training on each version separately, without the cross-input weight, achieves 58.7%. CrossRPO loses its gain over separate training when the cross-input weight comes from photos of other questions, so the gain depends on the contrast sets. ViRA-9B zooms onto the object as often as the agent trained on each version separately, but it reads the object correctly more often. ViRA-9B and ViRA-27B are the most accurate open agents of their sizes on ten and nine of the ten benchmarks we test, respectively. On ViRA-Bench, ViRA-27B also outperforms all proprietary models we test at every resolution (Figure 1b).

Contributions. Our contributions are as follows.

  • We propose CrossRPO, which trains on sets of inputs of any size. We prove that its update is unbiased, that it does not reinforce a set answered from common sense, and that the objectives of GRPO and of recent two-input methods are special cases of its objective.

  • We build ViRA-Train and ViRA-Bench, contrast sets made by editing one small object in photos of 21 to 120 MP, and ViRA-Bench-Real, photographed sets in which only the real object changes.

  • We train ViRA-9B and 27B with CrossRPO. Our experiments show that ViRA models outperform the models with similar size on most benchmarks. We will release all data, code and models.

Related Work

Zooming agents. Thinking-with-images agents [50, 60] crop and zoom over several turns. They are trained with supervised data [53] or with RL on the final answer [13, 25, 33, 59, 83, 85, 87]. Other methods search the image without training [35, 57, 67, 82] or learn to see fine details without zooming [70, 81]. Recent rewards credit crops that cover the evidence [11, 86] or calls that outperform a random crop or no call [14, 51], but they still judge one image at a time. Audits find that the returned crops often matter little [54, 69, 90] and that the gains come mostly from the model [47]. Except for PerceptRL [20], these works compare with proprietary models that view the image once and have no zoom tool [51, 59, 70, 71, 81], whereas we give them our zoom tool at their highest thinking level.

Learning from contrasts. Contrast sets and similar minimal pairs [15, 17, 34, 63] test whether answers follow the image, and they credit a model only if it answers every member correctly. Counterfactual examples also serve as supervised training data [7, 31, 62]. For RL, a reward on the final answer cannot distinguish an answer that is read from the image from an answer that is based on priors [42, 77, 79]. DeFacto [74] and EVAN [42] therefore add training signals from inputs without the evidence, whereas our versions replace the evidence, so every version has an answer to read. GRPO [21, 55] and its variants [2, 44, 80, 88] compare rollouts of one prompt, also for vision-language models [26, 45, 56, 68]. Recent methods train on augmented or rephrased copies of an input with the same answer [43, 76, 89, 91], on counterfactual pairs [12, 27, 37] or on packs of inputs [28, 30]. CRPO [12] weights an input and its transformed copy by the accuracy on the other, and CrossRPO extends this coupling to any number of inputs (Appendix § 8.11).

Cross-Input Relative Policy Optimization

On a single image, a correct guess earns the same reward as reading, so RL on final answers cannot distinguish them. A contrast set can distinguish them, because a guess is correct on only one of its versions. CrossRPO turns this difference into a training signal for visual agents. Figure 3 compares its pipeline with GRPO. The theorems, propositions and their proofs are in Appendix § 8.

Preliminaries

Preliminaries

Reward. The agent receives a question qq and an image ss downsampled to a pixel budget of xx MP. At each turn, it either answers or calls a zoom tool with a box on the image. The tool crops that region from the full-resolution image and resizes it to fit the same budget (Appendix § 10.§ 1). A rollout o∼Pθ(⋅∣s)o \sim P_\theta(\cdot\mid s) is the full sequence of turns on the input ss under the policy πθ\pi_\theta. For the correct answer yy, its correctness is c(o)=1[ans⁡(o)=y]c(o) = \mathbf{1}[\operatorname{ans}(o) = y], and its reward is r(o)=c(o)r(o) = c(o).

Group-based RL. CrossRPO builds on RL methods that compare several rollouts of the same input. These methods sample o1,…,on∼Pθ(⋅∣s)o_1,\ldots,o_n \sim P_\theta(\cdot\mid s) and update θ\theta along g^=∑iAi∇θlog⁡Pθ(oi∣s)\hat{g} = \sum_i A_i \nabla_\theta\log P_\theta(o_i \mid s), where the advantage AiA_i compares ci=c(oi)c_i = c(o_i) with the other rollouts. GRPO [55] uses Ai=(ci−μ)/σA_i = (c_i - \mu)/\sigma with μ=1n∑jcj\mu= \frac{1}{n}\sum_j c_j and σ2=1n∑j(cj−μ)2\sigma^2 = \frac{1}{n}\sum_j(c_j-\mu)^2, and Ai=0A_i = 0 if all cjc_j are equal. RLOO [2] uses Ai=ci−cˉ−iA_i = c_i - \bar{c}_{-i} with cˉ−i=1n−1∑j≠icj\bar{c}_{-i} = \frac{1}{n-1}\sum_{j\ne i} c_j. Both depend only on the rollouts of one input, so a correct guess and a correct reading receive the same advantage.

Contrast sets and the objective

A contrast set [15] is a tuple S=(q,(sv,yv)v=1K)S = (q, (s_v,y_v)_{v=1}^{K}) of a question qq and KK versions svs_v of an image with correct answers yvy_v. The versions are identical except in a small region with the evidence, and yu≠yvy_u \ne y_v for u≠vu \ne v. A single image is a set with K=1K = 1. The agent sees one version at a time. Let ov∼Pθ(⋅∣sv)o_v \sim P_\theta(\cdot\mid s_v) be independent rollouts, one on each version, with cv=1[ans⁡(ov)=yv]c_v = \mathbf{1}[\operatorname{ans}(o_v) = y_v]. The agent solves the set if c1=⋯=cK=1c_1 = \cdots= c_K = 1. The accuracy on version vv and the set accuracy are

pv=Pr⁡[cv=1],Pr⁡[c1=⋯=cK=1]=∏vpv.(1)p_v = \Pr[c_v = 1], \qquad\Pr[c_1 = \cdots= c_K = 1] = \prod_{v} p_v. \tag*{(1)}

Consider an agent that does not read the region that differs. Since the versions are identical outside this region, its answer has the same distribution on every version, ρ(a)=Pr⁡[ans⁡(ov)=a]\rho(a) = \Pr[\operatorname{ans}(o_v) = a] for all vv, and hence pv=ρ(yv)p_v = \rho(y_v). Since the yvy_v are pairwise different, ∑vpv=∑vρ(yv)≤∑aρ(a)≤1\sum_v p_v = \sum_v \rho(y_v) \le\sum_a \rho(a) \le1, and hence ∏vpv≤(1K∑vpv)K≤K−K\prod_v p_v \le\left(\frac{1}{K}\sum_v p_v\right)^K \le K^{-K} (Proposition A.11). If the agent always gives the same answer and K≥2K \ge2, then ∏vpv=0\prod_v p_v = 0. In contrast, an agent that reads can reach pv=1p_v = 1 for all vv and ∏vpv=1\prod_v p_v = 1. Set accuracy alone gives a weak signal early in training. The term of version vv in its gradient is ∏u≠vpu∇θpv\prod_{u\ne v} p_u \nabla_\theta p_v, which is zero when another version is never answered correctly. CrossRPO therefore maximizes the accuracy together with the set accuracy. For K≤3K \le3, which covers all our image sets, the objective is

J(θ)=∑v=1Kpv+λ∏v=1Kpv,λ≥0.(2)J(\theta) = \sum_{v=1}^{K} p_v + \lambda\prod_{v=1}^{K} p_v,\qquad\lambda\ge0. \tag*{(2)}

We call λ∏vpv\lambda\prod_v p_v the joint term, which rewards the agent for answering all versions correctly.

The CrossRPO advantage

References

  1. [1]Aishwarya Agrawal, Dhruv Batra, Devi Parikh, and Aniruddha Kembhavi. Don’t just assume; look and answer: Overcoming priors for visual question answering. In 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR, pages 4971–4980, Salt Lake City, UT, 2018.arxiv.org/abs/1712.00377
  2. [2]Arash Ahmadian, Chris Cremer, Matthias Gallé, Marzieh Fadaee, Julia Kreutzer, Olivier Pietquin, Ahmet Üstün, and Sara Hooker. Back to basics: Revisiting REINFORCE-style optimization for learning from human feedback in LLMs. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL, pages 12248–12267, Bangkok, Thailand, 2024.arxiv.org/abs/2402.14740
  3. [3]Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Ming-Hsuan 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. CoRR, abs/2502.13923, 2025.arxiv.org/abs/2502.13923
  4. [4]Baolong Bi, Shaohan Huang, Yiwei Wang, Tianchi Yang, Zihan Zhang, Haizhen Huang, Lingrui Mei, Junfeng Fang, Zhao Li, Furu Wei, Weiwei Deng, Feng Sun, Qi Zhang, and Shenghua Liu. Context-DPO: Aligning language models for context-faithfulness. In Findings of the Association for Computational Linguistics, ACL, pages 10280–10300, Vienna, Austria, 2025.
  5. [5]David Blackwell. Conditional expectation and unbiased sequential estimation. The Annals of Mathematical Statistics, 18(1):105–110, 1947.DOI
  6. [6]Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoubhik Debnath, Ronghang Hu, Didac Suris Coll-Vinent, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, Jie Lei, Tengyu Ma, Baishan Guo, Arpit Kalla, Markus Marks, Joseph Greer, Meng Wang, Peize Sun, Roman Rädle, Triantafyllos Afouras, Effrosyni Mavroudi, Katherine Xu, Tsung-Han Wu, Yu Zhou, Liliane Momeni, Rishi Hazra, Shuangrui Ding, Sagar Vaze, Francois Porcher, Feng Li, Siyuan Li, Aishwarya Kamath, Ho Kei Cheng, Piotr Dollár, Nikhila Ravi, Kate Saenko, Pengchuan Zhang, and Christoph Feichtenhofer. SAM 3: Segment anything with concepts. In The Fourteenth International Conference on Learning Representations, ICLR, Rio de Janeiro, Brazil, 2026.arxiv.org/abs/2511.16719
  7. [7]Long Chen, Xin Yan, Jun Xiao, Hanwang Zhang, Shiliang Pu, and Yueting Zhuang. Counterfactual samples synthesizing for robust visual question answering. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 10797–10806, Seattle, WA, 2020.arxiv.org/abs/2003.06576
  8. [8]Lin Chen, Jinsong Li, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Zehui Chen, Haodong Duan, Jiaqi Wang, Yu Qiao, Dahua Lin, and Feng Zhao. Are we on the right way for evaluating large vision-language models? In Advances in Neural Information Processing Systems, NeurIPS, Vancouver, Canada, 2024.arxiv.org/abs/2403.20330
  9. [9]Yong Xieng Chng, Tao Hu, Wenwen Tong, Xueheng Li, Jiandong Chen, Haojia Yu, Jiefan Lu, Hewei Guo, Hanming Deng, Xie Chen, Gao Huang, Dahua Lin, and Lewei Lu. SenseNova-MARS: Empowering multimodal agentic reasoning and search via reinforcement learning. CoRR, abs/2512.24330, 2025.
  10. [10]Lauro Langosco di Langosco, Jack Koch, Lee D. Sharkey, Jacob Pfau, and David Krueger. Goal misgeneralization in deep reinforcement learning. In International Conference on Machine Learning, ICML, pages 12004–12019, Baltimore, MD, 2022.
  11. [11]Qihua Dong, Gozde Sahin, Pei Wang, Zhaowei Cai, Robik Shrestha, Hao Yang, and Davide Modolo. Visual reasoning through tool-supervised reinforcement learning. CoRR, abs/2604.19945, 2026.arxiv.org/abs/2604.19945
  12. [12]Dazhao Du, Jian Liu, Jialong Qin, Tao Han, Bohai Gu, Fangqi Zhu, Yujia Zhang, Eric Liu, Xi Chen, and Song Guo. Learning spatiotemporal sensitivity in video LLMs via counterfactual reinforcement learning. CoRR, abs/2605.21988, 2026.arxiv.org/abs/2605.21988
  13. [13]Yue Fan, Xuehai He, Diji Yang, Kaizhi Zheng, Ching-Chen Kuo, Yuting Zheng, Xinze Guan, and Xin Wang. GRIT: Teaching MLLMs to think with images. In Advances in Neural Information Processing Systems, NeurIPS, San Diego, CA, 2025.
  14. [14]Mingkuan Feng, Jinyang Wu, Hao Gu, Fangrui Lv, Ruihan Jin, Chu Yuan Zhang, Zhengqi Wen, and Jianhua Tao. TACO: Tool-augmented credit optimization for agentic tool use. CoRR, abs/2606.30251, 2026.arxiv.org/abs/2606.30251
  15. [15]Matt Gardner, Yoav Artzi, Victoria Basmuova, Jonathan Berant, Ben Bogin, Sihao Chen, Pradeep Dasigi, Dheeru Dua, Yanai Elazar, Ananth Gottumukkala, Nitish Gupta, Hannah Hajishirzi, Gabriel Ilharco, Daniel Khashabi, Kevin Lin, Jiangming Liu, Nelson F. Liu, Phoebe Mulcaire, Qiang Ning, Sameer Singh, Noah A. Smith, Sanjay Subramanian, Reut Tsarfaty, Eric Wallace, Ally Zhang, and Ben Zhou. Evaluating models’ local decision boundaries via contrast sets. In Findings of the Association for Computational Linguistics: EMNLP, pages 1307–1323, Virtual, 2020.arxiv.org/abs/2004.02709
  16. [16]Robert Geirhos, Jörn-Henrik Jacobsen, Claudio Michaelis, Richard S. Zemel, Wieland Brendel, Matthias Bethge, and Felix A. Wichmann. Shortcut learning in deep neural networks. Nat. Mach. Intell., 2(11): 665–673, 2020.DOI
  17. [17]Yash Goyal, Tejas Khot, Douglas Summers-Stay, Dhruv Batra, and Devi Parikh. Making the V in VQA matter: Elevating the role of image understanding in visual question answering. In 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR, pages 6325–6334, Honolulu, HI, 2017.
  18. [18]Evan Greensmith, Peter L. Bartlett, and Jonathan Baxter. Variance reduction techniques for gradient estimates in reinforcement learning. J. Mach. Learn. Res., 5:1471–1530, 2004.DOI
  19. [19]Tianrui Guan, Fuxiao Liu, Xiyang Wu, Ruiqi Xian, Zongxia Li, Xiaoyu Liu, Xijun Wang, Lichang Chen, Furong Huang, Yaser Yacoob, Dinesh Manocha, and Tianyi Zhou. Hallusionbench: An advanced diagnostic suite for entangled language hallucination and visual illusion in large vision-language models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 14375–14385, Seattle, WA, 2024.
  20. [20]Tongkun Guan, Zhibo Yang, Mingkun Yang, Haocheng Wang, Wei Shen, and Xiaokang Yang. PerceptRL: Making visual perception rewardable in MLLMs from answers to interactions. arXiv preprint arXiv:2610.37471, 2026.arxiv.org/abs/2610.37471
  21. [21]Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Peiyi Wang, Qihao Zhu, Runxin Xu, Ruoyu Zhang, Shirong Ma, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, Bing Xue, Bingxuan Wang, Bochao Wu, Bei Feng, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chong Ruan, Damai Dai, Deli Chen, Dongjie Ji, Erhang Li, Fangyun Lin, Fucong Dai, Fuli Luo, Guangbo Hao, Guanting Chen, Guowei Li, Hao Zhang, Hanwei Xu, Honghui Ding, Huazuo Gao, Hui Qu, Hui Li, Jianzhong Guo, Jiashi Li, Jingchang Chen, Jingyang Yuan, Jinhao Tu, Junjie Qiu, Junlong Li, J. L. Cai, Jiaqi Ni, Jian Liang, Jin Chen, Kai Dong, Kai Hu, Kaichao You, Kaige Gao, Kang Guan, Kexin Huang, Kuai Yu, Lean Wang, Lecong Zhang, Liang Zhao, Litong Wang, Liyue Zhang, Lei Xu, Leyi Xia, Mingchuan Zhang, Minghua Zhang, Minghui Tang, Mingxu Zhou, Meng Li, Miaojun Wang, Mingming Li, Ning Tian, Panpan Huang, Peng Zhang, Qiancheng Wang, Qinyu Chen, Qiushi Du, Ruiqi Ge, Ruisong Zhang, Ruizhe Pan, Runji Wang, R. J. Chen, R. L. Jin, Ruyi Chen, Shanghao Lu, Shangyan Zhou, Shanhuang Chen, Shengfeng Ye, Shiyu Wang, Shuiping Yu, Shunfeng Zhou, Shuting Pan, S. S. Li, Shuang Zhou, Shaoqing Wu, Tao Yun, Tian Pei, Tianyu Sun, Tao Wang, Wangding Zeng, Wen Liu, Wenfeng Liang, Wenjun Gao, Wenqin Yu, Wentao Zhang, W. L. Xiao, Wei An, Xiaodong Liu, Xiaohan Wang, Xiaokang Chen, Xiaotao Nie, Xin Cheng, Xin Liu, Xin Xie, Xingchao Liu, Xinyu Yang, Xinyuan Li, Xuecheng Su, Xuheng Lin, X. Q. Li, Xiangyue Jin, Xiaojin Shen, Xiaosha Chen, Xiaowen Sun, Xiaoxiang Wang, Xinnan Song, Xinyi Zhou, Xianzu Wang, Xinxia Shan, Y. K. Li, Y. Q. Wang, Y. X. Wei, Yang Zhang, Yanhong Xu, Yao Li, Yao Zhao, Yaofeng Sun, Yaohui Wang, Yi Yu, Yichao Zhang, Yifan Shi, Yiliang Xiong, Ying He, Yishi Piao, Yisong Wang, Yixuan Tan, Yiyang Ma, Yiyuan Liu, Yongqiang Guo, Yuan Ou, Yuduan Wang, Yue Gong, Yuheng Zou, Yujia He, Yunfan Xiong, Yuxiang Luo, Yuxiang You, Yuxuan Liu, Yuyang Zhou, Y. X. Zhu, Yanping Huang, Yaohui Li, Yi Zheng, Yuchen Zhu, Yunxian Ma, Ying Tang, Yukun Zha, Yuting Yan, Z. Z. Ren, Zehui Ren, Zhangli Sha, Zhe Fu, Zhean Xu, Zhenda Xie, Zhengyan Zhang, Zhewen Hao, Zhicheng Ma, Zhigang Yan, Zhiyu Wu, Zhihui Gu, Zijia Zhu, Zijun Liu, Zilin Li, Ziwei Xie, Ziyang Song, Zizheng Pan, Zhen Huang, Zhipeng Xu, Zhongyu Zhang, and Zhen Zhang. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning. Nat., 645(8081):633–638, 2025.
  22. [22]Zonghao Guo, Ruyi Xu, Yuan Yao, Junbo Cui, Zanlin Ni, Chunjiang Ge, Tat-Seng Chua, Zhiyuan Liu, and Gao Huang. LLaVA-UHD: An LMM perceiving any aspect ratio and high-resolution images. In European Conference on Computer Vision, ECCV, pages 390–406, Milan, Italy, 2024.arxiv.org/abs/2403.11703
  23. [23]Godfrey Harold Hardy, John Edensor Littlewood, and George Pólya. Inequalities. Cambridge University Press, 1952.
  24. [24]Wassily Hoeffding. A class of statistics with asymptotically normal distribution. The Annals of Mathematical Statistics, 19(3):293–325, 1948.
  25. [25]Jack Hong, Chenxiao Zhao, ChengLin Zhu, Weiheng Lu, Guohai Xu, and Xing Yu. DeepEyeS2: Toward agentic multimodal model. In The Fourteenth International Conference on Learning Representations, ICLR, Rio de Janeiro, Brazil, 2026.arxiv.org/abs/2511.05271
  26. [26]Wenxuan Huang, Bohan Jia, Shaosheng Cao, Zheyu Ye, Fei Zhao, Zhe Xu, Yao Hu, and Shaohui Lin. Vision-R1: Incentivizing reasoning capability in multimodal large language models. In The Fourteenth International Conference on Learning Representations, ICLR, Rio de Janeiro, Brazil, 2026.arxiv.org/abs/2503.06749
  27. [27]Zhe Huang, Hao Wen, Aiming Hao, Bingze Song, Meiqi Wu, Jiahong Wu, Xiangxiang Chu, Sheng Lu, and Haoqian Wang. Taming hallucinations: Boosting MLLMs’ video understanding via counterfactual video generation. CoRR, abs/2512.24271, 2025.arxiv.org/abs/2512.24271
  28. [28]De Jiang, Zhengyang Zhang, Kehong Yuan, and Shaohua Ma. CARGO-VL: Counterfactual arbitration with risk-constrained group optimization for vision-language models. CoRR, abs/2608.04509, 2026.arxiv.org/abs/2608.04509
  29. [29]Bowen Jin, Hansi Zeng, Zhenrui Yue, Dong Wang, Hamed Zamani, and Jiawei Han. Search-R1: Training LLMs to reason and leverage search engines with reinforcement learning. CoRR, abs/2503.09516, 2025.arxiv.org/abs/2503.09516
  30. [30]Zhaolu Kang, Shiyu Liu, Tailong Luo, Wei Zhang, Yingjie He, Lei Wei, Guansu Wang, Liang He, Siheng Wang, Guangyuan Dong, et al. Behavior pack optimization for video MLLM post-training. arXiv preprint arXiv:2610.03141, 2026.arxiv.org/abs/2610.03141
  31. [31]Divyansh Kaushik, Eduard H. Hovy, and Zachary Chase Lipton. Learning the difference that makes A difference with counterfactually-augmented data. In 8th International Conference on Learning Representations, ICLR, Addis Ababa, Ethiopia, 2020.
  32. [32]Wouter Kool, Herke van Hoof, and Max Welling. Buy 4 REINFORCE samples, get a baseline for free! In Deep Reinforcement Learning Meets Structured Prediction, ICLR, Workshop, New Orleans, LA, 2019.
  33. [33]Xin Lai, Junyi Li, Wei Li, Tao Liu, Tianjian Li, and Hengshuang Zhao. Mini-o3: Scaling up reasoning patterns and interaction turns for visual search. In The Fourteenth International Conference on Learning Representations, ICLR, Rio de Janeiro, Brazil, 2026.arxiv.org/abs/2509.07969
  34. [34]Baiqi Li, Zhiqiu Lin, Wenxuan Peng, Jean de Dieu Nyandwi, Daniel Jiang, Zixian Ma, Simran Khanuja, Ranjay Krishna, Graham Neubig, and Deva Ramanan. NaturalBench: Evaluating vision-language models on natural adversarial samples. In Advances in Neural Information Processing Systems, NeurIPS, Vancouver, Canada, 2024.arxiv.org/abs/2410.14669
  35. [35]Geng Li, Jinglin Xu, Yunzhen Zhao, and Yuxin Peng. DyFo: A training-free dynamic focus visual search for enhancing LMMs in fine-grained visual understanding. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 9098–9108, Nashville, TN, 2025.arxiv.org/abs/2504.14920
  36. [36]Kaican Li, Lewei Yao, Jiannan Wu, Tiez heng Yu, Jierun Chen, Haoli Bai, Lu Hou, Lanqing Hong, Wei Zhang, and Nevin L. Zhang. InSight-03: Empowering multimodal foundation models with generalized visual search. In The Fourteenth International Conference on Learning Representations, ICLR, Rio de Janeiro, Brazil, 2026.
  37. [37]Suhang Li, Osamu Yoshie, and Yuya Ieiri. CFCamo: A counterfactual detect-or-abstain framework for camouflaged object detection. CoRR, abs/2606.11231, 2026.arxiv.org/abs/2606.11231
  38. [38]Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Wayne Xin Zhao, and Ji-Rong Wen. Evaluating object hallucination in large vision-language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP, pages 292–305, Singapore, 2023.arxiv.org/abs/2305.10355
  39. [39]Zhang Li, Biao Yang, Qiang Liu, Zhiyin Ma, Shuo Zhang, Jingxu Yang, Yabo Sun, Yuliang Liu, and Xiang Bai. Monkey: Image resolution and text label are important things for large multi-modal models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 26753–26763, Seattle, WA, 2024.arxiv.org/abs/2311.06607
  40. [40]Huawei Lin, Yunzhi Shi, Tong Geng, Weijie Zhao, Wei Wang, and Ravender Pal Singh. Agent-Omni: Test-time multimodal reasoning via model coordination for understanding anything. CoRR, abs/2511.02834, 2025.arxiv.org/abs/2511.02834
  41. [41]Huawei Lin, Peng Li, Jie Song, Fuxin Jiang, and Tieying Zhang. MUSE-Autoskill: Self-evolving agents via skill creation, memory, management, and evaluation. CoRR, abs/2605.27366, 2026.arxiv.org/abs/2605.27366
  42. [42]Chenxi Liu, Tianyi Xiong, Yanshuo Chen, and Heng Huang. Grounding multimodal reasoning with evidence-ablated negatives. In NeurIPS, 2026.arxiv.org/abs/2610.39237
  43. [43]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. In Advances in Neural Information Processing Systems, NeurIPS, San Diego, CA, 2025.arxiv.org/abs/2504.13055
  44. [44]Zichen Liu, Changyu Chen, Wenjun Li, Penghui Qi, Tianyu Pang, Chao Du, Wee Sun Lee, and Min Lin. Understanding R1-Zero-like training: A critical perspective. CoRR, abs/2503.20783, 2025.arxiv.org/abs/2503.20783
  45. [45]Ziyu Liu, Zeyi Sun, Yuhang Zhang, Xiaoyi Dong, Yuhang Cao, Haodong Duan, Dahua Lin, and Jiaqi Wang. Visual-RFT: Visual reinforcement fine-tuning. In IEEE/CVF International Conference on Computer Vision, ICCV, pages 2034–2044, Honolulu, HI, 2025.arxiv.org/abs/2503.01785
  46. [46]Shayne Longpre, Kartik Perisetla, Anthony Chen, Nikhil Ramesh, Chris DuBois, and Sameer Singh. Entity-based knowledge conflicts in question answering. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, EMNLP, pages 7052–7063, Punta Cana, Dominican Republic, 2021.
  47. [47]Yan Ma, Weiyu Zhang, Tianle Li, Linge Du, Xuyang Shen, and Pengfei Liu. What does vision tool-use reinforcement learning really learn? Disentangling tool-induced and intrinsic effects for crop-and-zoom. CoRR, abs/2602.01334, 2026.arxiv.org/abs/2602.01334
  48. [48]Yifei Ming, Senthil Purushwalkam, Shrey Pandit, Zixuan Ke, Xuan-Phi Nguyen, Caiming Xiong, and Shafiq Joty. FaithEval: Can your language model stay faithful to context, even if "the moon is made of marshmallows". In The Thirteenth International Conference on Learning Representations, ICLR, Singapore, 2025.arxiv.org/abs/2410.03727
  49. [49]Yulei Niu, Kaihua Tang, Hanwang Zhang, Zhiwu Lu, Xian-Sheng Hua, and Ji-Rong Wen. Counterfactual VQA: A cause-effect look at language bias. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR, pages 12700–12710, Virtual, 2021.arxiv.org/abs/2006.04315
  50. [50]OpenAI. Thinking with images. https://openai.com/index/thinking-with-images/, 2025. Blog post, April 16, 2025.
  51. [51]Kunyu Peng, Junming Liu, Ruiqi He, Qingzhuo Wang, Jianzhong Qi, and Xianhui Liu. When should a VLM look? Paying only for visual calls that were needed and used. CoRR, abs/2609.22910, 2026.arxiv.org/abs/2609.22910
  52. [52]John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. CoRR, abs/1707.06347, 2017.arxiv.org/abs/1707.06347
  53. [53]Hao Shao, Shengju Qian, Han Xiao, Guanglu Song, Zhuofan Zong, Letian Wang, Yu Liu, and Hongsheng Li. Visual CoT: Advancing multi-modal language models with a comprehensive dataset and benchmark for chain-of-thought reasoning. In Advances in Neural Information Processing Systems, NeurIPS, Vancouver, Canada, 2024.arxiv.org/abs/2403.16999
  54. [54]Jiahao Shao, Yuanbo Yang, Yiyi Liao, Yujun Shen, Ceyuan Yang, and Yinghao Xu. Thinking with tools, not with pixels: Tool calls as text scaffolds for visual reasoning. CoRR, abs/2608.09682, 2026.arxiv.org/abs/2608.09682
  55. [55]Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Mingchuan Zhang, Y. K. Li, Y. Wu, and Daya Guo. DeepSeekMath: Pushing the limits of mathematical reasoning in open language models. CoRR, abs/2402.03300, 2024.arxiv.org/abs/2402.03300
  56. [56]Haozhan Shen, Peng Liu, Jingcheng Li, Chunxin Fang, Yibo Ma, Jiajia Liao, Qiaoli Shen, Zilun Zhang, Kangjia Zhao, Qianqian Zhang, Ruochen Xu, and Tiancheng Zhao. VLM-R1: A stable and generalizable R1-style large vision-language model. CoRR, abs/2504.07615, 2025.arxiv.org/abs/2504.07615
  57. [57]Haozhan Shen, Kangjia Zhao, Tiancheng Zhao, Ruochen Xu, Zilun Zhang, Mingwei Zhu, and Jianwei Yin. ZoomEye: Enhancing multimodal LLMs with human-like zooming capabilities through tree-based image exploration. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP, pages 6602–6618, Suzhou, China, 2025.arxiv.org/abs/2411.16044
  58. [58]Guangming Sheng, Chi Zhang, Zilingfeng Ye, Xibin Wu, Wang Zhang, Ru Zhang, Yanghua Peng, Haibin Lin, and Chuan Wu. HybridFlow: A flexible and efficient RLHF framework. In Proceedings of the Twentieth European Conference on Computer Systems, EuroSys, pages 1279–1297, Rotterdam, The Netherlands, 2025.DOI
  59. [59]Alex Su, Haozhe Wang, Weiming Ren, Fangzhen Lin, and Wenhu Chen. Pixel Reasoner: Incentivizing pixel space reasoning via curiosity-driven reinforcement learning. In Advances in Neural Information Processing Systems, NeurIPS, San Diego, CA, 2025.DOI
  60. [60]Zhaoc​​hen Su, Peng Xiang, Hangyu Guo, Zhenhua Liu, Yan Ma, Xiaoye Qu, Jiaqi Liu, Yanshu Li, Kaide Zeng, Zhengyuan Yang, Linjie Li, Yu Cheng, Heng Ji, Junxian He, and Yi R. (May) Fung. Thinking with images for multimodal reasoning: Foundations, methods, and future frontiers. CoRR, abs/2506.23918, 2025.arxiv.org/abs/2506.23918
  61. [61]Qwen Team. Qwen3-VL technical report. CoRR, abs/2511.21631, 2025.
  62. [62]Damien Teney, Ehsan Abbasnejad, and Anton van den Hengel. Learning what makes a difference from counterfactual examples and gradient supervision. In European Conference on Computer Vision, ECCV, pages 580–599, Glasgow, UK, 2020.arxiv.org/abs/2004.09034
  63. [63]Tristan Thrush, Ryan Jiang, Max Bartolo, Amanpreet Singh, Adina Williams, Douwe Kiela, and Candace Ross. Winoground: Probing vision and language models for visio-linguistic compositionality. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 5228–5238, New Orleans, LA, 2022.arxiv.org/abs/2204.03162
  64. [64]Shengbang Tong, Zhuang Liu, Yuexiang Zhai, Yi Ma, Yann LeCun, and Saining Xie. Eyes wide shut? Exploring the visual shortcomings of multimodal LLMs. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 9568–9578, Seattle, WA, 2024.arxiv.org/abs/2401.06209
  65. [65]Haochen Wang, Xiangtai Li, Zilong Huang, Anran Wang, Jiacong Wang, Tao Zhang, Jiani Zheng, Sule Bai, Zijian Kang, Jiashi Feng, Zhuochen Wang, and Zhaoxiang Zhang. Traceable evidence enhanced visual grounded reasoning: Evaluation and method. In The Fourteenth International Conference on Learning Representations, ICLR, Rio de Janeiro, Brazil, 2026.
  66. [66]Wenbin Wang, Liang Ding, Minyan Zeng, Xiabin Zhou, Li Shen, Yong Luo, Wei Yu, and Dacheng Tao. Divide, conquer and combine: A training-free framework for high-resolution image perception in multimodal large language models. In Thirty-Ninth AAAI Conference on Artificial Intelligence, Thirty-Seventh Conference on Innovative Applications of Artificial Intelligence, Fifteenth Symposium on Educational Advances in Artificial Intelligence, AAAI, pages 7907–7915, Philadelphia, PA, 2025.arxiv.org/abs/2408.15556
  67. [67]Wenbin Wang, Yongcheng Jing, Liang Ding, Yingjie Wang, Li Shen, Yong Luo, Bo Du, and Dacheng Tao. Retrieval-augmented perception: High-resolution image perception meets visual RAG. In Forty-second International Conference on Machine Learning, ICML, Vancouver, Canada, 2025.arxiv.org/abs/2503.01222
  68. [68]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. In Advances in Neural Information Processing Systems, NeurIPS, San Diego, CA, 2025.arxiv.org/abs/2504.07934
  69. [69]Zhiheng Wang, Bo Peng, Lai Wei, and Chaochao Lu. The illusion of visual tool-use: A causal audit of thinking with images. CoRR, abs/2608.06270, 2026.arxiv.org/abs/2608.06270
  70. [70]Lai Wei, Liangbo He, Jun Lan, Lingzhong Dong, Yutong Cai, Siyuan Li, Huijia Zhu, Weiqiang Wang, Linghe Kong, Yue Wang, Zhuosheng Zhang, and Weiran Huang. Zooming without zooming: Region-to-image distillation for fine-grained multimodal perception. CoRR, abs/2602.11858, 2026.arxiv.org/abs/2602.11858
  71. [71]Lai Wei, Yatai Li, Liangbo He, Jun Lan, Renjie Gu, Linhan Cao, Junkai Lin, Huijia Zhu, Weiqiang Wang, Linghe Kong, et al. Zooming without zooming v2: From seeing details to connecting evidence for generalizable multimodal perception. arXiv preprint arXiv:2610.37037, 2026.arxiv.org/abs/2610.37037
  72. [72]Ronald J. Williams. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Mach. Learn., 8:229–256, 1992.
  73. [73]Penghao Wu and Saining Xie. V*: Guided visual search as a core mechanism in multimodal LLMs. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 13084–13094, Seattle, WA, 2024.arxiv.org/abs/2312.14135
  74. [74]Tianrun Xu, Haoda Jing, Ye Li, Yuquan Wei, Jun Feng, Guanyu Chen, Haichuan Gao, Tianren Zhang, and Feng Chen. Defacto: Counterfactual thinking with images for enforcing evidence-grounded and faithful reasoning. CoRR, abs/2509.20912, 2025.arxiv.org/abs/2509.20912
  75. [75]Zhongxing Xu, Chengzhi Liu, Qingyue Wei, Juncheng Wu, James Y. Zou, Xin Wang, Yuyin Zhou, and Sheng Liu. More thinking, less seeing? Assessing amplified hallucination in multimodal reasoning models. In Advances in Neural Information Processing Systems, NeurIPS, San Diego, CA, 2025.arxiv.org/abs/2505.21523
  76. [76]Huanjin Yao, Qixiang Yin, Jingyi Zhang, Min Yang, Yibo Wang, Wenhao Wu, Fei Su, Li Shen, Minghui Qiu, Dacheng Tao, and Jiaxing Huang. R1-ShareVL: Incentivizing reasoning capabilities of multimodal large language models via Share-GRPO. In Advances in Neural Information Processing Systems, NeurIPS, San Diego, CA, 2025.DOI
  77. [77]Jiayu Yao, Yiwei Wang, Anmeng Zhang, Zhe Sun, Songsong Wang, Lingrui Mei, Yuyao Ge, and Shenghua Liu. Multimodal reward hacking in reinforcement learning. CoRR, abs/2607.09492, 2026.arxiv.org/abs/2607.09492
  78. [78]Ruilin Yao, Xinyu Qiu, Tianyu Zhou, Yi Rong, Shengwu Xiong, Shuheng Shen, and Changhua Meng. Learning to look beyond screens: Transferable visual observation from GUI grounding. arXiv preprint arXiv:2610.37093, 2026.arxiv.org/abs/2610.37093
  79. [79]Haocun Ye, Xinlong Jiang, Qile Chen, Bingyu Wang, Teng Zhang, Shubai Chen, Tingyu Wu, Zhenkun Zheng, and Yiqiang Chen. Same reward, different skills: When multimodal RL learns to look. arXiv preprint arXiv:2610.01908, 2026.arxiv.org/abs/2610.01908
  80. [80]Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Juncai Liu, Lingjun Liu, Xin Liu, Haibin Lin, Zhiqi Lin, Bole Ma, Guangming Sheng, Yuxuan Tong, Chi Zhang, Mofan Zhang, Ru Zhang, Wang Zhang, Hang Zhu, Jinhua Zhu, Jiaze Chen, Jiangjie Chen, Chengyi Wang, Hongli Yu, Yuxuan Song, Xiangpeng Wei, Hao Zhou, Jingjing Liu, Wei-Ying Ma, Ya-Qin Zhang, Lin Yan, Yonghui Wu, and Mingxuan Wang. DAPO: An open-source LLM reinforcement learning system at scale. In Advances in Neural Information Processing Systems, NeurIPS, San Diego, CA, 2025.arxiv.org/abs/2503.14476
  81. [81]Qianhao Yuan, Jie Lou, Xing Yu, Hongyu Lin, Le Sun, Xianpei Han, and Yaojie Lu. Vision-OPD: Learning to see fine details for multimodal LLMs via on-policy self-distillation. CoRR, abs/2605.18740, 2026.arxiv.org/abs/2605.18740
  82. [82]Jiarui Zhang, Mahyar Khayatkhoei, Prateek Chhikara, and Filip Ilievski. MLLMs know where to look: Training-free perception of small visual details with multimodal LLMs. In The Thirteenth International Conference on Learning Representations, ICLR, Singapore, 2025.arxiv.org/abs/2502.17422
  83. [83]Xintong Zhang, Zhi Gao, Bofei Zhang, Pengxiang Li, Xiaowen Zhang, Yang Liu, Tao Yuan, Yuwei Wu, Yunde Jia, Song-Chun Zhu, and Qing Li. Chain-of-Focus: Adaptive visual search and zooming for multimodal reasoning via RL. CoRR, abs/2505.15436, 2025.arxiv.org/abs/2505.15436
  84. [84]Yifan Zhang, Huanyu Zhang, Haochen Tian, Chaoyou Fu, Shuangqing Zhang, Junfei Wu, Feng Li, Kun Wang, Qingsong Wen, Zhang Zhang, Liang Wang, and Rong Jin. MME-RealWorld: Could your multimodal LLM challenge high-resolution real-world scenarios that are difficult for humans? In The Thirteenth International Conference on Learning Representations, ICLR, Singapore, 2025.arxiv.org/abs/2408.13257
  85. [85]Yifan Zhang, Xingyu Lu, Shukang Yin, Chaoyou Fu, Wei Chen, Xiao Hu, Bin Wen, Kaiyu Jiang, Changyi Liu, Tianke Zhang, Haonan Fan, Kaibing Chen, Jiankang Chen, Haojie Ding, Kaiyu Tang, Zhang Zhang, Liang Wang, Fan Yang, Tingting Gao, and Guorui Zhou. Thyme: Think beyond images. In The Fourteenth International Conference on Learning Representations, ICLR, Rio de Janeiro, Brazil, 2026.arxiv.org/abs/2508.11630
  86. [86]Xuanpu Zhao, Zhentao Tan, Dianmo Sheng, Tianxiang Chen, Yao Liu, Yue Wu, Tao Gong, Qi Chu, and Nenghai Yu. Learning to focus and precise cropping: A reinforcement learning framework with information gaps and grounding loss for MLLMs. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 25938–25947, Denver, CO, 2026.arxiv.org/abs/2603.27494
  87. [87]Ziwei Zheng, Minghao Yang, Jack Hong, Chenxiao Zhao, Guohai Xu, Le Yang, Chao Shen, and Xing Yu. DeepEyes: Incentivizing "thinking with images" via reinforcement learning. In The Fourteenth International Conference on Learning Representations, ICLR, Rio de Janeiro, Brazil, 2026.arxiv.org/abs/2505.14362
  88. [88]Hongyi Zhou, Kai Ye, Erhan Xu, Jin Zhu, Ying Yang, Shijin Gong, and Chengchun Shi. Demystifying group relative policy optimization: Its policy gradient is a U-statistic. CoRR, abs/2603.01162, 2026.arxiv.org/abs/2603.01162
  89. [89]Chenghua Zhu, Zhaolu Kang, Qifan Shi, Siyan Wu, Kehan Jiang, Lei Wei, Lianyu Hu, Guangyuan Dong, Mingbo Yang, Rui Lu, and Guibo Luo. CoMET: Contrastive motion-enhanced temporal reasoning for video multimodal large language models. CoRR, abs/2608.21030, 2026.arxiv.org/abs/2608.21030
  90. [90]Didi Zhu, Changrui Chen, Stefanos Zafeiriou, and Jiankang Deng. VisualFLIP: Do predictions depend on task-critical visual evidence in multimodal reasoning? CoRR, abs/2606.07872, 2026.arxiv.org/abs/2606.07872
  91. [91]Xin Zou, Haolin Deng, Yibo Yan, Shuliang Liu, Kening Zheng, Zhiwei Jin, Chen Chen, Haonan Lu, and Xuming Hu. Consistency as inductive bias: Learning cross-view invariance for robust multimodal reasoning. CoRR, abs/2606.29812, 2026.arxiv.org/abs/2606.29812

Paper details

vlnc:W_we8zam5w
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

LLMs, image editing models, Claude Code, Codex

View statement
Looking for related papers…
Contents