The following papers were recommended by the Semantic Scholar API
\n- \n
- ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning (2025) \n
- Ariadne: A Controllable Framework for Probing and Extending VLM Reasoning Boundaries (2025) \n
- V-Thinker: Interactive Thinking with Images (2025) \n
- Long Grounded Thoughts: Distilling Compositional Visual Reasoning Chains at Scale (2025) \n
- Efficient Reasoning via Thought-Training and Thought-Free Inference (2025) \n
- Artemis: Structured Visual Reasoning for Perception Policy Learning (2025) \n
- CodeV: Code with Images for Faithful Visual Reasoning via Tool-Aware Policy Optimization (2025) \n
Please give a thumbs up to this comment if you found it helpful!
\nIf you want recommendations for any Paper on Hugging Face checkout this Space
\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend
Great work, thansk a lot!
\n","updatedAt":"2025-12-04T06:10:58.689Z","author":{"_id":"652b7fc6756a15d750fcba7b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/652b7fc6756a15d750fcba7b/B92dB4HruUXtBvTbQi3lD.jpeg","fullname":"Dongyuan Li","name":"coffee3699","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.714651346206665},"editors":["coffee3699"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/652b7fc6756a15d750fcba7b/B92dB4HruUXtBvTbQi3lD.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2511.22586","authors":[{"_id":"692d22f24397b1ec214f690b","name":"Yifan Du","hidden":false},{"_id":"692d22f24397b1ec214f690c","name":"Kun Zhou","hidden":false},{"_id":"692d22f24397b1ec214f690d","name":"Yingqian Min","hidden":false},{"_id":"692d22f24397b1ec214f690e","name":"Yue Ling","hidden":false},{"_id":"692d22f24397b1ec214f690f","name":"Wayne Xin Zhao","hidden":false},{"_id":"692d22f24397b1ec214f6910","name":"Youbin Wu","hidden":false}],"publishedAt":"2025-11-27T16:19:34.000Z","submittedOnDailyAt":"2025-12-03T00:27:02.051Z","title":"Revisiting the Necessity of Lengthy Chain-of-Thought in Vision-centric Reasoning Generalization","submittedOnDailyBy":{"_id":"61d78857a21a9b49c7e8e4a9","avatarUrl":"/avatars/c7e7f84cad775be2d13fab8530bf21f5.svg","isPro":false,"fullname":"Yifan Du","user":"Richard1999","type":"user"},"summary":"We study how different Chain-of-Thought (CoT) designs affect the acquisition of the generalizable visual reasoning ability in vision-language models (VLMs). While CoT data, especially long or visual CoT such as \"think with image\", has been widely used to supervise intermediate reasoning, it remains unclear why specific CoT designs help and which ones truly support generalizable reasoning. To systematically evaluate this, we focus on a controlled maze-solving benchmark where reasoning rules are fully visual, difficulty can be tuned by grid size, and all the intermediate steps can be automatically generated. Using Qwen2.5-VL-7B under a standard SFT-then-RL pipeline, we compare three representative CoT formats: Language CoT, Grounding CoT (with spatial coordinate trajectories), and Visual CoT (with image manipulations). Our experiments reveal that visual and longer CoT mainly accelerate convergence but do not lift the final performance ceiling; concise CoT containing only essential grounding steps outperforms longer traces; and, strikingly, CoT retaining only the minimal grounding results generalizes best across different maze sizes. We further validate these insights on other vision-centric tasks. These findings highlight a \"short is long\" effect and provide practical guidance for constructing more generalizable SFT datasets for visual reasoning.","upvotes":7,"discussionId":"692d22f24397b1ec214f6911","projectPage":"https://github.com/RUCAIBox/Revisiting-Visual-CoT","githubRepo":"https://github.com/Richar-Du/Revisiting-Visual-CoT","githubRepoAddedBy":"user","ai_summary":"Investigating different Chain-of-Thought designs in vision-language models reveals that concise grounding steps are most effective for improving generalizable visual reasoning across various tasks.","ai_keywords":["Chain-of-Thought","vision-language models","visual reasoning","CoT","Grounding CoT","Visual CoT","maze-solving benchmark","Qwen2.5-VL-7B","SFT-then-RL","visual reasoning tasks"],"githubStars":2,"organization":{"_id":"67d1140985ea0644e2f14b99","name":"ByteDance-Seed","fullname":"ByteDance Seed","avatar":"https://cdn-uploads.huggingface.co/production/uploads/6535c9e88bde2fae19b6fb25/flkDUqd_YEuFsjeNET3r-.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"61d78857a21a9b49c7e8e4a9","avatarUrl":"/avatars/c7e7f84cad775be2d13fab8530bf21f5.svg","isPro":false,"fullname":"Yifan Du","user":"Richard1999","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"6342796a0875f2c99cfd313b","avatarUrl":"/avatars/98575092404c4197b20c929a6499a015.svg","isPro":false,"fullname":"Yuseung \"Phillip\" Lee","user":"phillipinseoul","type":"user"},{"_id":"646b43deb1202bc77c1024a4","avatarUrl":"/avatars/cf791574ab986bac274e7fbcf04e2a59.svg","isPro":false,"fullname":"hangyu guo","user":"Rosiness","type":"user"},{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user"},{"_id":"67ad9767d3a5cc6789882e10","avatarUrl":"/avatars/e5d16dc828670963ffe3bf2cf33318ab.svg","isPro":false,"fullname":"Kim Jiwan","user":"JiwanKim","type":"user"},{"_id":"686db5d4af2b856fabbf13aa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/6BjMv2LVNoqvbX8fQSTPI.png","isPro":false,"fullname":"V bbbb","user":"Bbbbbnnn","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0,"organization":{"_id":"67d1140985ea0644e2f14b99","name":"ByteDance-Seed","fullname":"ByteDance Seed","avatar":"https://cdn-uploads.huggingface.co/production/uploads/6535c9e88bde2fae19b6fb25/flkDUqd_YEuFsjeNET3r-.png"}}">Revisiting the Necessity of Lengthy Chain-of-Thought in Vision-centric Reasoning Generalization
Abstract
Investigating different Chain-of-Thought designs in vision-language models reveals that concise grounding steps are most effective for improving generalizable visual reasoning across various tasks.
We study how different Chain-of-Thought (CoT) designs affect the acquisition of the generalizable visual reasoning ability in vision-language models (VLMs). While CoT data, especially long or visual CoT such as "think with image", has been widely used to supervise intermediate reasoning, it remains unclear why specific CoT designs help and which ones truly support generalizable reasoning. To systematically evaluate this, we focus on a controlled maze-solving benchmark where reasoning rules are fully visual, difficulty can be tuned by grid size, and all the intermediate steps can be automatically generated. Using Qwen2.5-VL-7B under a standard SFT-then-RL pipeline, we compare three representative CoT formats: Language CoT, Grounding CoT (with spatial coordinate trajectories), and Visual CoT (with image manipulations). Our experiments reveal that visual and longer CoT mainly accelerate convergence but do not lift the final performance ceiling; concise CoT containing only essential grounding steps outperforms longer traces; and, strikingly, CoT retaining only the minimal grounding results generalizes best across different maze sizes. We further validate these insights on other vision-centric tasks. These findings highlight a "short is long" effect and provide practical guidance for constructing more generalizable SFT datasets for visual reasoning.
Community
We study how different Chain-of-Thought (CoT) designs affect the acquisition of the generalizable visual reasoning ability in vision-language models (VLMs). While CoT data, especially long or visual CoT such as "think with image", has been widely used to supervise intermediate reasoning, it remains unclear why specific CoT designs help and which ones truly support generalizable reasoning. To systematically evaluate this, we focus on a controlled maze-solving benchmark where reasoning rules are fully visual, difficulty can be tuned by grid size, and all the intermediate steps can be automatically generated. Using Qwen2.5-VL-7B under a standard SFT-then-RL pipeline, we compare three representative CoT formats: Language CoT, Grounding CoT (with spatial coordinate trajectories), and Visual CoT (with image manipulations). Our experiments reveal that visual and longer CoT mainly accelerate convergence but do not lift the final performance ceiling; concise CoT containing only essential grounding steps outperforms longer traces; and, strikingly, CoT retaining only the minimal grounding results generalizes best across different maze sizes. We further validate these insights on other vision-centric tasks. These findings highlight a "short is long" effect and provide practical guidance for constructing more generalizable SFT datasets for visual reasoning.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning (2025)
- Ariadne: A Controllable Framework for Probing and Extending VLM Reasoning Boundaries (2025)
- V-Thinker: Interactive Thinking with Images (2025)
- Long Grounded Thoughts: Distilling Compositional Visual Reasoning Chains at Scale (2025)
- Efficient Reasoning via Thought-Training and Thought-Free Inference (2025)
- Artemis: Structured Visual Reasoning for Perception Policy Learning (2025)
- CodeV: Code with Images for Faithful Visual Reasoning via Tool-Aware Policy Optimization (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Great work, thansk a lot!
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper