\nState-of-the-art optical flow method without any task-specific biases.","html":"<img src=\"https://cdn-uploads.huggingface.co/production/uploads/663692c75f67f8da32723bf8/usG_8dFrPsjc4ALvDVYTf.png\" width=\"500\" alt=\"FreeFlow\">\nState-of-the-art optical flow method without any task-specific biases.","updatedAt":"2026-09-11T09:16:32.316Z","author":{"_id":"663692c75f67f8da32723bf8","avatarUrl":"/avatars/258264afe2ea5048a4a7a8e9945d2f5b.svg","fullname":"Alexander Yakovenko","name":"a-yakovenko","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.5209192633628845},"editors":["a-yakovenko"],"editorAvatarUrls":["/avatars/258264afe2ea5048a4a7a8e9945d2f5b.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.11486","authors":[{"_id":"6aa3706347a406da7901e753","name":"Vladislav Bargatin","hidden":false},{"_id":"6aa3706347a406da7901e754","user":{"_id":"663692c75f67f8da32723bf8","avatarUrl":"/avatars/258264afe2ea5048a4a7a8e9945d2f5b.svg","isPro":false,"fullname":"Alexander Yakovenko","user":"a-yakovenko","type":"user","name":"a-yakovenko"},"name":"Alexander Yakovenko","status":"claimed_verified","statusLastChangedAt":"2026-09-11T09:53:20.868Z","hidden":false},{"_id":"6aa3706347a406da7901e755","name":"Khaled Abud","hidden":false},{"_id":"6aa3706347a406da7901e756","name":"Dmitriy Vatolin","hidden":false}],"publishedAt":"2026-09-10T00:00:00.000Z","submittedOnDailyAt":"2026-09-11T00:00:00.000Z","title":"FreeFlow: A Bias-free Hierarchical Transformer for Optical Flow Estimation","submittedOnDailyBy":{"_id":"663692c75f67f8da32723bf8","avatarUrl":"/avatars/258264afe2ea5048a4a7a8e9945d2f5b.svg","isPro":false,"fullname":"Alexander Yakovenko","user":"a-yakovenko","type":"user","name":"a-yakovenko"},"summary":"Optical flow methods typically rely on task-specific inductive biases, such as correlation volumes, feature warping, and iterative refinement, among others, to reach high accuracy. While effective, such biases constrain the model to predefined heuristics, which can limit its expressivity and lead to more complex pipelines and additional computational cost. We present FreeFlow, a hierarchical transformer built without any flow-specific components, using instead a single feed-forward encoder--decoder. FreeFlow combines three attention variants: window attention for local processing, shifted-window attention for cross-window information exchange, and a global attention operating at a reduced resolution. The resulting architecture scales naturally with model capacity, enabling a consistent accuracy gain from small to large variants. Despite the absence of standard inductive biases, FreeFlow achieves state-of-the-art results on major benchmarks, including Sintel (0.68/1.48 EPE on Clean/Final), KITTI-2015 (3.23 Fl-all), and Spring (3.192 1px), while remaining memory efficient at 1080p inference.","upvotes":5,"discussionId":"6aa3706347a406da7901e757","projectPage":"https://github.com/msu-video-group/freeflow","githubRepo":"https://github.com/msu-video-group/freeflow","githubRepoAddedBy":"user","ai_summary":"FreeFlow is a hierarchical transformer for optical flow that eliminates task-specific inductive biases and achieves state-of-the-art accuracy using window, shifted-window, and global attention.","ai_keywords":["optical flow","correlation volumes","feature warping","iterative refinement","hierarchical transformer","feed-forward encoder-decoder","window attention","shifted-window attention","global attention","Sintel","KITTI-2015","Spring"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":2},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"663692c75f67f8da32723bf8","avatarUrl":"/avatars/258264afe2ea5048a4a7a8e9945d2f5b.svg","isPro":false,"fullname":"Alexander Yakovenko","user":"a-yakovenko","type":"user"},{"_id":"623a3b20fa4890c51b04cba7","avatarUrl":"/avatars/ebc71284fab04d506471aa5b1792157b.svg","isPro":false,"fullname":"Andrey Moskalenko","user":"ANDRYHA","type":"user"},{"_id":"67f029c89d26e9aed8c3a79c","avatarUrl":"/avatars/f881bc3178a9b8d644bd2c8fa94a90d2.svg","isPro":false,"fullname":"Vladislav Pyatov","user":"VladPyatov","type":"user"},{"_id":"644aa54d1acffad9353d5655","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/644aa54d1acffad9353d5655/KE2nbf6UsmdNMFRtthwlH.jpeg","isPro":false,"fullname":"Sergey Kushneryuk","user":"skushneryuk","type":"user"},{"_id":"6a2da6c8ca070ee12c6e396c","avatarUrl":"/avatars/0355287dcabaa67dbc7f0b10b87451f9.svg","isPro":false,"fullname":"Joe Mama","user":"JoeMama123123123","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.11486.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2609.11486 in a dataset README.md to link it from this page.