Excited to share MeshFlow — a new approach that can generate meshes with a fraction of seconds, while achieving state-of-the-art generation quality.<br>Secret sources? Instead of autoregressive models, use equivariant flow-matching!<br>Code and pretrained checkpoints are ready! We’ll present MeshFlow next month at SIGGRAPH 2026 in LA.</p>\n","updatedAt":"2026-06-23T06:48:14.461Z","author":{"_id":"667fb15d2698e0647180089d","avatarUrl":"/avatars/d550c82668a3819ce23706c7fdf19475.svg","fullname":"Qi Sun","name":"qsun2001","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.8648263812065125},"editors":["qsun2001"],"editorAvatarUrls":["/avatars/d550c82668a3819ce23706c7fdf19475.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.23489","authors":[{"_id":"6a3a28b3fdcd3514343bb6d5","name":"Qi Sun","hidden":false},{"_id":"6a3a28b3fdcd3514343bb6d6","name":"Kiyohiro Nakayama","hidden":false},{"_id":"6a3a28b3fdcd3514343bb6d7","name":"Jing Nathan Yan","hidden":false},{"_id":"6a3a28b3fdcd3514343bb6d8","name":"Qixing Huang","hidden":false},{"_id":"6a3a28b3fdcd3514343bb6d9","name":"Alexander Rush","hidden":false},{"_id":"6a3a28b3fdcd3514343bb6da","name":"Leonidas Guibas","hidden":false},{"_id":"6a3a28b3fdcd3514343bb6db","name":"Gordon Wetzstein","hidden":false},{"_id":"6a3a28b3fdcd3514343bb6dc","name":"Jing Liao","hidden":false},{"_id":"6a3a28b3fdcd3514343bb6dd","name":"Guandao Yang","hidden":false}],"publishedAt":"2026-06-22T00:00:00.000Z","submittedOnDailyAt":"2026-06-23T00:00:00.000Z","title":"MeshFlow: Mesh Generation with Equivariant Flow Matching","submittedOnDailyBy":{"_id":"667fb15d2698e0647180089d","avatarUrl":"/avatars/d550c82668a3819ce23706c7fdf19475.svg","isPro":false,"fullname":"Qi Sun","user":"qsun2001","type":"user","name":"qsun2001"},"summary":"Meshes are among the most common 3D scene representations, but directly generating meshes is challenging because the representation contains important symmetries, including permutation invariance of faces and vertices. MeshFlow learns to generate triangle meshes directly as triangle soups, avoiding the need to serialize meshes into long autoregressive sequences. We adopt equivariant optimal-transport flow matching models that respect the key symmetries of triangle soups: arbitrary permutations of faces and permutations of the vertices within each face.\n Toward this goal, we propose a simple yet effective modification to the Diffusion Transformer architecture, resulting in a scalable network capable of modeling a velocity field while maintaining the desired equivariance. We further introduce an optimal-transport-based training objective that improves convergence by eliminating supervision signals that violate these symmetries. MeshFlow achieves mesh quality comparable to state-of-the-art autoregressive mesh generators while providing about an 18times speedup during inference. Project page is at https://qiisun.github.io/MeshFlow/.","upvotes":2,"discussionId":"6a3a28b3fdcd3514343bb6de","projectPage":"https://qiisun.github.io/MeshFlow/","githubRepo":"https://github.com/qiisun/MeshFlow","githubRepoAddedBy":"user","ai_summary":"MeshFlow generates triangle meshes directly using equivariant optimal-transport flow matching models with improved inference speed over autoregressive methods.","ai_keywords":["mesh generation","triangle soups","permutation invariance","equivariant optimal-transport flow matching","Diffusion Transformer","velocity field","optimal-transport-based training objective"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":27},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"667fb15d2698e0647180089d","avatarUrl":"/avatars/d550c82668a3819ce23706c7fdf19475.svg","isPro":false,"fullname":"Qi Sun","user":"qsun2001","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.23489.md","query":{}}">
MeshFlow: Mesh Generation with Equivariant Flow Matching
Published on Jun 22
· Submitted by Qi Sun on Jun 23 Abstract
MeshFlow generates triangle meshes directly using equivariant optimal-transport flow matching models with improved inference speed over autoregressive methods.
Meshes are among the most common 3D scene representations, but directly generating meshes is challenging because the representation contains important symmetries, including permutation invariance of faces and vertices. MeshFlow learns to generate triangle meshes directly as triangle soups, avoiding the need to serialize meshes into long autoregressive sequences. We adopt equivariant optimal-transport flow matching models that respect the key symmetries of triangle soups: arbitrary permutations of faces and permutations of the vertices within each face.
Toward this goal, we propose a simple yet effective modification to the Diffusion Transformer architecture, resulting in a scalable network capable of modeling a velocity field while maintaining the desired equivariance. We further introduce an optimal-transport-based training objective that improves convergence by eliminating supervision signals that violate these symmetries. MeshFlow achieves mesh quality comparable to state-of-the-art autoregressive mesh generators while providing about an 18times speedup during inference. Project page is at https://qiisun.github.io/MeshFlow/.
Community
Excited to share MeshFlow — a new approach that can generate meshes with a fraction of seconds, while achieving state-of-the-art generation quality.
Secret sources? Instead of autoregressive models, use equivariant flow-matching!
Code and pretrained checkpoints are ready! We’ll present MeshFlow next month at SIGGRAPH 2026 in LA.
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/2606.23489 in a model README.md to link it from this page.
Cite arxiv.org/abs/2606.23489 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2606.23489 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.