Project page：<a href=\"https://zry000.github.io/Canvas360/\" rel=\"nofollow\">https://zry000.github.io/Canvas360/</a><br>GitHub repo：<a href=\"https://github.com/Insta360-Research-Team/Canvas360\" rel=\"nofollow\">https://github.com/Insta360-Research-Team/Canvas360</a></p>\n","updatedAt":"2026-07-10T06:41:34.520Z","author":{"_id":"65240d0ca801972b6eb12ed8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65240d0ca801972b6eb12ed8/hl2RAssBperb5JlgOIDvw.jpeg","fullname":"Haoran Feng","name":"fenghora","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"zh","probability":0.3591594099998474},"editors":["fenghora"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/65240d0ca801972b6eb12ed8/hl2RAssBperb5JlgOIDvw.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.08765","authors":[{"_id":"6a50936375fd3d966bd45e88","name":"Haoran Feng","hidden":false},{"_id":"6a50936375fd3d966bd45e89","name":"Ruiyang Zhang","hidden":false},{"_id":"6a50936375fd3d966bd45e8a","name":"Longyi Zhang","hidden":false},{"_id":"6a50936375fd3d966bd45e8b","name":"Dizhe Zhang","hidden":false},{"_id":"6a50936375fd3d966bd45e8c","name":"Lu Qi","hidden":false}],"publishedAt":"2026-07-09T00:00:00.000Z","submittedOnDailyAt":"2026-07-10T00:00:00.000Z","title":"Enhancing In-context Panoramic Generation via Geometric-aware Pretraining","submittedOnDailyBy":{"_id":"65240d0ca801972b6eb12ed8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65240d0ca801972b6eb12ed8/hl2RAssBperb5JlgOIDvw.jpeg","isPro":false,"fullname":"Haoran Feng","user":"fenghora","type":"user","name":"fenghora"},"summary":"In this work, we present Canvas360, a two-stage framework for in-context panoramic generation that combines geometry-aware pretraining with downstream task-specific fine-tuning. To address the lack of large-scale, high-quality training data tailored to in-context panoramic tasks, we propose Canvas360Dataset, a collection of 1M high-quality paired panoramic samples for style transfer, inpainting, outpainting, and editing, enabling effective supervision across diverse in-context generation scenarios. On the modeling side, Canvas360 enhances text-to-panorama generation through parallel depth generation, velocity circular padding, and similarity loss regularization, enabling the model to learn geometry-aware representations, capture object distortion details, and improve geometric consistency and global coherence. Furthermore, empowered by strong panoramic priors, Canvas360 enables a unified in-context panoramic generation framework that supports diverse downstream tasks via token-level concatenation, surpassing prior methods in both task coverage and modeling flexibility. Extensive experiments show that Canvas360 improves panoramic image fidelity, achieving particularly strong performance on the panorama-specific FAED metric and competitive or leading results across the reported quantitative evaluations. More information can be found on our project page: https://zry000.github.io/Canvas360/","upvotes":14,"discussionId":"6a50936475fd3d966bd45e8d","projectPage":"https://zry000.github.io/Canvas360/","githubRepo":"https://github.com/Insta360-Research-Team/Canvas360","githubRepoAddedBy":"user","ai_summary":"Canvas360 is a two-stage framework for in-context panoramic generation that combines geometry-aware pretraining with fine-tuning, featuring a large-scale dataset and novel modeling techniques for improved geometric consistency and global coherence.","ai_keywords":["in-context panoramic generation","geometry-aware pretraining","downstream task-specific fine-tuning","panoramic image fidelity","parallel depth generation","velocity circular padding","similarity loss regularization","geometric consistency","global coherence","token-level concatenation","panorama-specific FAED metric"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":5,"organization":{"_id":"644035437663594a1263c7ff","name":"Insta360","fullname":"Insta360","avatar":"https://www.gravatar.com/avatar/1af68c704803d53d59ca8a5d8a9bc3e8?d=retro&size=100"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"65240d0ca801972b6eb12ed8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65240d0ca801972b6eb12ed8/hl2RAssBperb5JlgOIDvw.jpeg","isPro":false,"fullname":"Haoran Feng","user":"fenghora","type":"user"},{"_id":"6812e92b59113438668d0d76","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/YoH9F13mxXeydN7NetXLz.png","isPro":false,"fullname":"Derek Zhang","user":"derekzzzz","type":"user"},{"_id":"67fa181d7cfcb309de076b01","avatarUrl":"/avatars/42d613a5dd0d86949dd9d7fdff57e7cc.svg","isPro":false,"fullname":"Liu","user":"Richrq","type":"user"},{"_id":"68b69f1dae7231fbc481ecca","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/68b69f1dae7231fbc481ecca/-XKbUzjRnfPE9gA-PiDI6.png","isPro":false,"fullname":"HuXing","user":"CeFf1z","type":"user"},{"_id":"66e3d4ed9d477f566f7172a6","avatarUrl":"/avatars/b33ccbed744a1c582418a680eeed8962.svg","isPro":false,"fullname":"Zxkai","user":"kaikai2001","type":"user"},{"_id":"6805c0947c5fa8020f5ae11a","avatarUrl":"/avatars/71667c70dd8b9ca916b8811922d2d043.svg","isPro":false,"fullname":"sujia wang","user":"wsj0103","type":"user"},{"_id":"66fe2ed1a77ec468fff148f8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66fe2ed1a77ec468fff148f8/mpl-gYiGq0V36IXLgAko2.jpeg","isPro":false,"fullname":"Quanzhu Niu","user":"QuanzhuNiu","type":"user"},{"_id":"6790a04a1c897e8d8fc531e0","avatarUrl":"/avatars/e40422c89bf6d25467e9d6ed3292063c.svg","isPro":false,"fullname":"Mao","user":"AMao12138","type":"user"},{"_id":"6166e77ae4e97c73a070398d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6166e77ae4e97c73a070398d/trstFxS-ev9Zu2bPyaznh.jpeg","isPro":false,"fullname":"马琦","user":"Asimok","type":"user"},{"_id":"69f1cb405b28ceb06c92a782","avatarUrl":"/avatars/85cc668b03b6eebc7f546737c099381f.svg","isPro":false,"fullname":"weiying","user":"wing1212","type":"user"},{"_id":"67da87d1020ee588d62bde5e","avatarUrl":"/avatars/17402c63da6ccb0660c3bc1e2675f182.svg","isPro":false,"fullname":"Li","user":"JackCV","type":"user"},{"_id":"6a2625ab41d82ed47a3db055","avatarUrl":"/avatars/e6da22017a4aa1df2b48ad6f9c5d4366.svg","isPro":false,"fullname":"xpydgqb","user":"xpydgqb","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"644035437663594a1263c7ff","name":"Insta360","fullname":"Insta360","avatar":"https://www.gravatar.com/avatar/1af68c704803d53d59ca8a5d8a9bc3e8?d=retro&size=100"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.08765.md","query":{}}">
Enhancing In-context Panoramic Generation via Geometric-aware Pretraining
Abstract
Canvas360 is a two-stage framework for in-context panoramic generation that combines geometry-aware pretraining with fine-tuning, featuring a large-scale dataset and novel modeling techniques for improved geometric consistency and global coherence.
In this work, we present Canvas360, a two-stage framework for in-context panoramic generation that combines geometry-aware pretraining with downstream task-specific fine-tuning. To address the lack of large-scale, high-quality training data tailored to in-context panoramic tasks, we propose Canvas360Dataset, a collection of 1M high-quality paired panoramic samples for style transfer, inpainting, outpainting, and editing, enabling effective supervision across diverse in-context generation scenarios. On the modeling side, Canvas360 enhances text-to-panorama generation through parallel depth generation, velocity circular padding, and similarity loss regularization, enabling the model to learn geometry-aware representations, capture object distortion details, and improve geometric consistency and global coherence. Furthermore, empowered by strong panoramic priors, Canvas360 enables a unified in-context panoramic generation framework that supports diverse downstream tasks via token-level concatenation, surpassing prior methods in both task coverage and modeling flexibility. Extensive experiments show that Canvas360 improves panoramic image fidelity, achieving particularly strong performance on the panorama-specific FAED metric and competitive or leading results across the reported quantitative evaluations. More information can be found on our project page: https://zry000.github.io/Canvas360/
Community
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2607.08765 in a model README.md to link it from this page.
Cite arxiv.org/abs/2607.08765 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2607.08765 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.