Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

One Diffusion Step to Real-World Super-Resolution via Flow Trajectory Distillation (2025)
Optimizing for the Shortest Path in Denoising Diffusion Model (2025)
TraFlow: Trajectory Distillation on Pre-Trained Rectified Flow (2025)
Raccoon: Multi-stage Diffusion Training with Coarse-to-Fine Curating Videos (2025)
VipDiff: Towards Coherent and Diverse Video Inpainting via Training-free Denoising Diffusion Models (2025)
MaskGWM: A Generalizable Driving World Model with Video Mask Reconstruction (2025)
CustomVideoX: 3D Reference Attention Driven Dynamic Adaptation for Zero-Shot Customized Video Diffusion Transformers (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: \n\n@librarian-bot\n\t recommend

\n","updatedAt":"2025-03-11T01:34:32.532Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6933835744857788},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2503.04824","authors":[{"_id":"67cec02929ee2ca420c0c395","name":"Lei Ke","hidden":false},{"_id":"67cec02929ee2ca420c0c396","name":"Haohang Xu","hidden":false},{"_id":"67cec02929ee2ca420c0c397","name":"Xuefei Ning","hidden":false},{"_id":"67cec02929ee2ca420c0c398","name":"Yu Li","hidden":false},{"_id":"67cec02929ee2ca420c0c399","name":"Jiajun Li","hidden":false},{"_id":"67cec02929ee2ca420c0c39a","user":{"_id":"650be23ec4e52db6a4db63ef","avatarUrl":"/avatars/03af548029b38bee49ec295fefe74f9a.svg","isPro":false,"fullname":"Haoling Li","user":"Ringo1110","type":"user"},"name":"Haoling Li","status":"claimed_verified","statusLastChangedAt":"2025-03-10T11:08:46.234Z","hidden":false},{"_id":"67cec02929ee2ca420c0c39b","name":"Yuxuan Lin","hidden":false},{"_id":"67cec02929ee2ca420c0c39c","name":"Dongsheng Jiang","hidden":false},{"_id":"67cec02929ee2ca420c0c39d","name":"Yujiu Yang","hidden":false},{"_id":"67cec02929ee2ca420c0c39e","user":{"_id":"642ec9831d1737803dc1c30a","avatarUrl":"/avatars/c9ded838bad09004c15a27200e66a108.svg","isPro":false,"fullname":"linfeng zhang","user":"linfengZ","type":"user"},"name":"Linfeng Zhang","status":"claimed_verified","statusLastChangedAt":"2025-03-25T08:21:37.134Z","hidden":false}],"publishedAt":"2025-03-05T04:50:53.000Z","submittedOnDailyAt":"2025-03-10T09:04:36.587Z","title":"ProReflow: Progressive Reflow with Decomposed Velocity","submittedOnDailyBy":{"_id":"650be23ec4e52db6a4db63ef","avatarUrl":"/avatars/03af548029b38bee49ec295fefe74f9a.svg","isPro":false,"fullname":"Haoling Li","user":"Ringo1110","type":"user"},"summary":"Diffusion models have achieved significant progress in both image and video\ngeneration while still suffering from huge computation costs. As an effective\nsolution, flow matching aims to reflow the diffusion process of diffusion\nmodels into a straight line for a few-step and even one-step generation.\nHowever, in this paper, we suggest that the original training pipeline of flow\nmatching is not optimal and introduce two techniques to improve it. Firstly, we\nintroduce progressive reflow, which progressively reflows the diffusion models\nin local timesteps until the whole diffusion progresses, reducing the\ndifficulty of flow matching. Second, we introduce aligned v-prediction, which\nhighlights the importance of direction matching in flow matching over magnitude\nmatching. Experimental results on SDv1.5 and SDXL demonstrate the effectiveness\nof our method, for example, conducting on SDv1.5 achieves an FID of 10.70 on\nMSCOCO2014 validation set with only 4 sampling steps, close to our teacher\nmodel (32 DDIM steps, FID = 10.05).","upvotes":9,"discussionId":"67cec02b29ee2ca420c0c447","ai_summary":"Flow matching is improved with progressive reflow and aligned v-prediction, reducing computational costs while maintaining high image generation quality.","ai_keywords":["diffusion models","flow matching","progressive reflow","aligned v-prediction","FID","MSCOCO2014","sampling steps","DDIM steps"]},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"650be23ec4e52db6a4db63ef","avatarUrl":"/avatars/03af548029b38bee49ec295fefe74f9a.svg","isPro":false,"fullname":"Haoling Li","user":"Ringo1110","type":"user"},{"_id":"6553316bf151de82f6a23e1d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6553316bf151de82f6a23e1d/GTBkSj4Fa3OoyM6Muz_Sc.jpeg","isPro":false,"fullname":"Gongye Liu","user":"liuhuohuo","type":"user"},{"_id":"651ed7ef755e92f7f12742e6","avatarUrl":"/avatars/57a9cc189b4a59299aad6c96191b18d8.svg","isPro":false,"fullname":"yu li","user":"lyabc","type":"user"},{"_id":"659fe17d58a49686b2b4aae9","avatarUrl":"/avatars/4163bb79967e92efd0a0d9af26441fb1.svg","isPro":false,"fullname":"kl","user":"kl233","type":"user"},{"_id":"648eb1eb59c4e5c87dc116e0","avatarUrl":"/avatars/c636cea39c2c0937f01398c94ead5dad.svg","isPro":false,"fullname":"fdsqefsgergd","user":"T-representer","type":"user"},{"_id":"66f612b934b8ac9ffa44f084","avatarUrl":"/avatars/6836c122e19c66c90f1673f28b30d7f0.svg","isPro":false,"fullname":"Tang","user":"tommysally","type":"user"},{"_id":"675d0066c6b100978f143484","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/675d0066c6b100978f143484/TRyTJph2QWFfyKh-WlSX6.jpeg","isPro":false,"fullname":"Mia Scott","user":"darkPhoenix09","type":"user"},{"_id":"6720d9fd72358ddfaf90c42e","avatarUrl":"/avatars/f7a06b8d131af375b0c62c4d35c47b6d.svg","isPro":false,"fullname":"GM","user":"MgGladys","type":"user"},{"_id":"6351e5bb3734c6e8a5c1bec1","avatarUrl":"/avatars/a784a51b369b197398575c3afbd5ceab.svg","isPro":false,"fullname":"Han-Bit Kang","user":"hbkang","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0}">

Papers

arxiv:2503.04824

ProReflow: Progressive Reflow with Decomposed Velocity

Published on Mar 5, 2025

· Submitted by

Haoling Li on Mar 10, 2025

Upvote

Authors:

Haoling Li ,

Linfeng Zhang

Abstract

Flow matching is improved with progressive reflow and aligned v-prediction, reducing computational costs while maintaining high image generation quality.

AI-generated summary

Diffusion models have achieved significant progress in both image and video generation while still suffering from huge computation costs. As an effective solution, flow matching aims to reflow the diffusion process of diffusion models into a straight line for a few-step and even one-step generation. However, in this paper, we suggest that the original training pipeline of flow matching is not optimal and introduce two techniques to improve it. Firstly, we introduce progressive reflow, which progressively reflows the diffusion models in local timesteps until the whole diffusion progresses, reducing the difficulty of flow matching. Second, we introduce aligned v-prediction, which highlights the importance of direction matching in flow matching over magnitude matching. Experimental results on SDv1.5 and SDXL demonstrate the effectiveness of our method, for example, conducting on SDv1.5 achieves an FID of 10.70 on MSCOCO2014 validation set with only 4 sampling steps, close to our teacher model (32 DDIM steps, FID = 10.05).