Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456 Paper page - Scale-wise Distillation of Diffusion Models
\n","updatedAt":"2025-03-21T21:52:35.996Z","author":{"_id":"63151e278d85ad332fa4b905","avatarUrl":"/avatars/d00c42a6862e6335dbccab68cb63e2c0.svg","fullname":"Jeremy","name":"jbaron34","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6489382386207581},"editors":["jbaron34"],"editorAvatarUrls":["/avatars/d00c42a6862e6335dbccab68cb63e2c0.svg"],"reactions":[],"isReport":false},"replies":[{"id":"67deb307f38795c54546e44c","author":{"_id":"62b6cc49752323892323bc04","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62b6cc49752323892323bc04/gGBld1KJIP9AIpd81L3PC.jpeg","fullname":"Dmitry Baranchuk","name":"dbaranchuk","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":29,"isUserFollowing":false},"createdAt":"2025-03-22T12:54:31.000Z","type":"comment","data":{"edited":true,"hidden":false,"latest":{"raw":"Thank you so much for pointing this out! Indeed, there was a bug in the inference code related to the upscaling method. After fixing it, the aliasing artifacts became negligible (take a look at the images). We truly appreciate you highlighting this issue. Feel free to try out the demo and share your feedback :)\n\n\n\n\n","html":"
Thank you so much for pointing this out! Indeed, there was a bug in the inference code related to the upscaling method. After fixing it, the aliasing artifacts became negligible (take a look at the images). We truly appreciate you highlighting this issue. Feel free to try out the demo and share your feedback :)
\n
\n","updatedAt":"2025-03-22T13:09:50.404Z","author":{"_id":"62b6cc49752323892323bc04","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62b6cc49752323892323bc04/gGBld1KJIP9AIpd81L3PC.jpeg","fullname":"Dmitry Baranchuk","name":"dbaranchuk","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":29,"isUserFollowing":false}},"numEdits":2,"identifiedLanguage":{"language":"en","probability":0.7283640503883362},"editors":["dbaranchuk"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/62b6cc49752323892323bc04/gGBld1KJIP9AIpd81L3PC.jpeg"],"reactions":[{"reaction":"❤️","users":["not-lain","rherrmann","hunoutl","bmay"],"count":4}],"isReport":false,"parentCommentId":"67dddfa3679c7d6dee3818d8"}}]},{"id":"67de1394f25418b0534567c0","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},"createdAt":"2025-03-22T01:34:12.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [Learning Few-Step Diffusion Models by Trajectory Distribution Matching](https://huggingface.co/papers/2503.06674) (2025)\n* [Adding Additional Control to One-Step Diffusion with Joint Distribution Matching](https://huggingface.co/papers/2503.06652) (2025)\n* [SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillation](https://huggingface.co/papers/2503.09641) (2025)\n* [Accelerate High-Quality Diffusion Models with Inner Loop Feedback](https://huggingface.co/papers/2501.13107) (2025)\n* [One-Step Residual Shifting Diffusion for Image Super-Resolution via Distillation](https://huggingface.co/papers/2503.13358) (2025)\n* [One-Step Diffusion Model for Image Motion-Deblurring](https://huggingface.co/papers/2503.06537) (2025)\n* [Denoising Score Distillation: From Noisy Diffusion Pretraining to One-Step High-Quality Generation](https://huggingface.co/papers/2503.07578) (2025)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
\n
The following papers were recommended by the Semantic Scholar API
Please give a thumbs up to this comment if you found it helpful!
\n
If 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
\n","updatedAt":"2025-03-22T01:34:12.805Z","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":0.7017001509666443},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2503.16397","authors":[{"_id":"67dd1227046f2c38458e9588","user":{"_id":"6410d3a4cfbe9c4400233d1e","avatarUrl":"/avatars/9e1c4e0ea5f6964c90013f6e19c41db1.svg","isPro":false,"fullname":"nikita","user":"quickjkee","type":"user"},"name":"Nikita Starodubcev","status":"claimed_verified","statusLastChangedAt":"2025-03-24T08:53:21.020Z","hidden":false},{"_id":"67dd1227046f2c38458e9589","user":{"_id":"629cf0475a13ba8233dd18c9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1654452258405-noauth.jpeg","isPro":false,"fullname":"Denis Kuznedelev","user":"SpiridonSunRotator","type":"user"},"name":"Denis Kuznedelev","status":"admin_assigned","statusLastChangedAt":"2025-03-21T12:59:13.018Z","hidden":false},{"_id":"67dd1227046f2c38458e958a","name":"Artem Babenko","hidden":false},{"_id":"67dd1227046f2c38458e958b","user":{"_id":"62b6cc49752323892323bc04","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62b6cc49752323892323bc04/gGBld1KJIP9AIpd81L3PC.jpeg","isPro":false,"fullname":"Dmitry Baranchuk","user":"dbaranchuk","type":"user"},"name":"Dmitry Baranchuk","status":"admin_assigned","statusLastChangedAt":"2025-03-21T12:59:23.033Z","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/6410d3a4cfbe9c4400233d1e/lkzM32YzNPrhP9ESkaUpF.png"],"publishedAt":"2025-03-20T17:54:02.000Z","submittedOnDailyAt":"2025-03-21T08:05:39.256Z","title":"Scale-wise Distillation of Diffusion Models","submittedOnDailyBy":{"_id":"6410d3a4cfbe9c4400233d1e","avatarUrl":"/avatars/9e1c4e0ea5f6964c90013f6e19c41db1.svg","isPro":false,"fullname":"nikita","user":"quickjkee","type":"user"},"summary":"We present SwD, a scale-wise distillation framework for diffusion models\n(DMs), which effectively employs next-scale prediction ideas for\ndiffusion-based few-step generators. In more detail, SwD is inspired by the\nrecent insights relating diffusion processes to the implicit spectral\nautoregression. We suppose that DMs can initiate generation at lower data\nresolutions and gradually upscale the samples at each denoising step without\nloss in performance while significantly reducing computational costs. SwD\nnaturally integrates this idea into existing diffusion distillation methods\nbased on distribution matching. Also, we enrich the family of distribution\nmatching approaches by introducing a novel patch loss enforcing finer-grained\nsimilarity to the target distribution. When applied to state-of-the-art\ntext-to-image diffusion models, SwD approaches the inference times of two full\nresolution steps and significantly outperforms the counterparts under the same\ncomputation budget, as evidenced by automated metrics and human preference\nstudies.","upvotes":41,"discussionId":"67dd1229046f2c38458e9617","projectPage":"https://yandex-research.github.io/swd/","githubRepo":"https://github.com/yandex-research/swd","githubRepoAddedBy":"user","ai_summary":"A scale-wise distillation framework for diffusion models reduces computational costs and improves inference times by incorporating next-scale predictions and enhancing distribution matching methods.","ai_keywords":["diffusion models","scale-wise distillation","next-scale prediction","implicit spectral autoregression","denoising","computational costs","distribution matching","patch loss","inference times","text-to-image diffusion models"],"githubStars":113},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6410d3a4cfbe9c4400233d1e","avatarUrl":"/avatars/9e1c4e0ea5f6964c90013f6e19c41db1.svg","isPro":false,"fullname":"nikita","user":"quickjkee","type":"user"},{"_id":"629cf0475a13ba8233dd18c9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1654452258405-noauth.jpeg","isPro":false,"fullname":"Denis Kuznedelev","user":"SpiridonSunRotator","type":"user"},{"_id":"67d145625e5bc6c00f3556bd","avatarUrl":"/avatars/c7a27485d21f7ee155cfeaacd07691a4.svg","isPro":false,"fullname":"Aleksei","user":"LeonMeon","type":"user"},{"_id":"64381042819f3ab20d1779d6","avatarUrl":"/avatars/36fcd43693992d4b0fe8f3a141be35bb.svg","isPro":false,"fullname":"Eremeev Dmitriy","user":"eremeev-d","type":"user"},{"_id":"655f8492e3dda95b340822ae","avatarUrl":"/avatars/69e2bed8045faf35105e15f228db0567.svg","isPro":false,"fullname":"George","user":"MrDarkTesla","type":"user"},{"_id":"630bd0e9910e17bbfeadf22c","avatarUrl":"/avatars/ff041e7a8cbca8e7130c3102a5d31c0c.svg","isPro":false,"fullname":"Ivan Rubachev","user":"puhsu","type":"user"},{"_id":"6240eecb585605a407953360","avatarUrl":"/avatars/8517adfe7ade40faca1afa7791749c66.svg","isPro":false,"fullname":"Zmushko Philip","user":"fzmushko","type":"user"},{"_id":"64aeb6aa59d35c5f8180ba7c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64aeb6aa59d35c5f8180ba7c/udqEsHBhzjZl_ShvgjD5D.jpeg","isPro":false,"fullname":"Sergey Kastryulin","user":"snk4tr","type":"user"},{"_id":"6261af8040e04009e813a43d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6261af8040e04009e813a43d/cBuawglGStA12f4DxHtRo.jpeg","isPro":false,"fullname":"Roman Garipov","user":"garipovroma","type":"user"},{"_id":"62b6cc49752323892323bc04","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62b6cc49752323892323bc04/gGBld1KJIP9AIpd81L3PC.jpeg","isPro":false,"fullname":"Dmitry Baranchuk","user":"dbaranchuk","type":"user"},{"_id":"6655ae06ec19a307d34f8764","avatarUrl":"/avatars/fbab83d3f7667e1c96954824976678f4.svg","isPro":false,"fullname":"Yury Gorishniy","user":"Yura52","type":"user"},{"_id":"64a42977250bfdecd9570a9e","avatarUrl":"/avatars/df5d7cf159e6bb9e961e1c77d1b89d36.svg","isPro":false,"fullname":"Daniil Selikhanovych","user":"apryc1","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0}">
A scale-wise distillation framework for diffusion models reduces computational costs and improves inference times by incorporating next-scale predictions and enhancing distribution matching methods.
AI-generated summary
We present SwD, a scale-wise distillation framework for diffusion models
(DMs), which effectively employs next-scale prediction ideas for
diffusion-based few-step generators. In more detail, SwD is inspired by the
recent insights relating diffusion processes to the implicit spectral
autoregression. We suppose that DMs can initiate generation at lower data
resolutions and gradually upscale the samples at each denoising step without
loss in performance while significantly reducing computational costs. SwD
naturally integrates this idea into existing diffusion distillation methods
based on distribution matching. Also, we enrich the family of distribution
matching approaches by introducing a novel patch loss enforcing finer-grained
similarity to the target distribution. When applied to state-of-the-art
text-to-image diffusion models, SwD approaches the inference times of two full
resolution steps and significantly outperforms the counterparts under the same
computation budget, as evidenced by automated metrics and human preference
studies.
Scale-wise Distillation (SwD) is a novel framework for accelerating diffusion models (DMs) by progressively increasing spatial resolution during the generation process. SwD achieves significant speedups (2.5× to 10×) compared to full-resolution models while maintaining or even improving image quality.
Tried it out and noticed that it struggles with aliasing artifacts due to the upscaling. Have you tried any alternative interpolation methods on the upscale step?
Thank you so much for pointing this out! Indeed, there was a bug in the inference code related to the upscaling method. After fixing it, the aliasing artifacts became negligible (take a look at the images). We truly appreciate you highlighting this issue. Feel free to try out the demo and share your feedback :)