{"id":15046,"date":"2023-05-29T20:40:56","date_gmt":"2023-05-29T12:40:56","guid":{"rendered":"http:\/\/139.9.1.231\/?p=15046"},"modified":"2023-07-10T16:45:58","modified_gmt":"2023-07-10T08:45:58","slug":"stable-diffusion","status":"publish","type":"post","link":"http:\/\/139.9.1.231\/index.php\/2023\/05\/29\/stable-diffusion\/","title":{"rendered":"stable diffusion\uff1a\u6f5c\u5728\u6269\u6563\u6a21\u578b"},"content":{"rendered":"\n<p>\u53c2\u8003\uff1a<\/p>\n\n\n\n<p>1\u3001<a href=\"https:\/\/zhuanlan.zhihu.com\/p\/573984443\">https:\/\/zhuanlan.zhihu.com\/p\/573984443<\/a><\/p>\n\n\n\n<p>2\u3001<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\">https:\/\/zhangzhenhu.github.io\/blog\/aigc<\/a><\/p>\n\n\n\n<p>3\u3001 <a href=\"https:\/\/zhuanlan.zhihu.com\/p\/599160988\">https:\/\/zhuanlan.zhihu.com\/p\/599160988<\/a> <\/p>\n\n\n\n<h2>\u6269\u6563\u6982\u7387\u6a21\u578b\uff08diffusion probabilistic models\uff09<\/h2>\n\n\n\n<p>1\u3001&nbsp;\u6269\u6563\u6982\u7387\u6a21\u578b\uff08diffusion probabilistic model\uff09<\/p>\n\n\n\n<p>2\u3001\u964d\u566a\u6269\u6563\u6982\u7387\u6a21\u578b\uff08Denoising diffusion probabilistic model,DDPM\uff09<\/p>\n\n\n\n<p>3\u3001\u57fa\u4e8e\u5206\u6570\u7684\u89e3\u91ca\uff08Score-based DDPM\uff09<\/p>\n\n\n\n<p>4\u3001\u6269\u6563\u6a21\u578b\u7684\u4e09\u79cd\u7b49\u4ef7\u8868\u793a<\/p>\n\n\n\n<p>5\u3001\u6539\u8fdb\u964d\u566a\u6269\u6563\u6982\u7387\u6a21\u578b\uff08Improved Denoising Diffusion Probabilistic Models,IDDPM\uff09<\/p>\n\n\n\n<p>6.&nbsp;\u53c2\u8003\u6587\u732e<\/p>\n\n\n\n<p>Jascha Sohl-Dickstein, Eric&nbsp;A. Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. 2015.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/1503.03585\">arXiv:1503.03585<\/a>.2<em>(<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id3\">1<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id5\">2<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id7\">3<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id11\">4<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id12\">5<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id13\">6<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id21\">7<\/a>)<\/em><\/p>\n\n\n\n<p>Calvin Luo. Understanding diffusion models: a unified perspective. 2022.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2208.11970\">arXiv:2208.11970<\/a>.3<em>(<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id9\">1<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id15\">2<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id16\">3<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id17\">4<\/a>)<\/em><\/p>\n\n\n\n<p>Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. 2020.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2006.11239\">arXiv:2006.11239<\/a>.<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id10\">4<\/a><\/p>\n\n\n\n<p>Diederik&nbsp;P. Kingma, Tim Salimans, Ben Poole, and Jonathan Ho. Variational diffusion models. 2022.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2107.00630\">arXiv:2107.00630<\/a>.<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/%E6%89%A9%E6%95%A3%E6%A6%82%E7%8E%87%E6%A8%A1%E5%9E%8B.html#id19\">5<\/a><\/p>\n\n\n\n<p>Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. 2019.\u00a0<a href=\"https:\/\/arxiv.org\/abs\/1907.05600\">arXiv:1907.05600<\/a>.<\/p>\n\n\n\n<h2>\u53bb\u566a\u6269\u6563\u9690\u5f0f\u6a21\u578b\uff08Denoising Diffusion Implicit Models,DDIM\uff09<\/h2>\n\n\n\n<p>Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. 2022.\u00a0<a href=\"https:\/\/arxiv.org\/abs\/2010.02502\">arXiv:2010.02502<\/a>.<\/p>\n\n\n\n<h2>\u57fa\u4e8e\u5206\u6570\u7684\u751f\u6210\u6a21\u578b<\/h2>\n\n\n\n<p>Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. 2019.\u00a0<a href=\"https:\/\/arxiv.org\/abs\/1907.05600\">arXiv:1907.05600<\/a>.<\/p>\n\n\n\n<p>Yang Song, Jascha Sohl-Dickstein, Diederik\u00a0P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. 2021.\u00a0<a href=\"https:\/\/arxiv.org\/abs\/2011.13456\">arXiv:2011.13456<\/a>.<\/p>\n\n\n\n<p>Aapo Hyv\u00e4rinen and Peter Dayan. Estimation of non-normalized statistical models by score matching.\u00a0<em>Journal of Machine Learning Research<\/em>, 2005.<\/p>\n\n\n\n<p>Yang Song and Stefano Ermon. Improved techniques for training score-based generative models. 2020.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2006.09011\">arXiv:2006.09011<\/a>.<\/p>\n\n\n\n<h2>\u6761\u4ef6\u63a7\u5236\u6269\u6563\u6a21\u578b<\/h2>\n\n\n\n<p>Prafulla Dhariwal and Alex Nichol. Diffusion models beat gans on image synthesis. 2021.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2105.05233\">arXiv:2105.05233<\/a>.2<em>(<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/Guidance.html#id3\">1<\/a>,<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/Guidance.html#id5\">2<\/a>)<\/em><\/p>\n\n\n\n<p>Calvin Luo. Understanding diffusion models: a unified perspective. 2022.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2208.11970\">arXiv:2208.11970<\/a>.<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/Guidance.html#id6\">3<\/a><\/p>\n\n\n\n<p>Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. 2022.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2207.12598\">arXiv:2207.12598<\/a>.<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/Guidance.html#id7\">4<\/a><\/p>\n\n\n\n<p>Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. Glide: towards photorealistic image generation and editing with text-guided diffusion models. 2022.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2112.10741\">arXiv:2112.10741<\/a>.<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/Guidance.html#id8\">5<\/a><\/p>\n\n\n\n<p>Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. 2022.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2204.06125\">arXiv:2204.06125<\/a>.<a href=\"https:\/\/zhangzhenhu.github.io\/blog\/aigc\/Guidance.html#id9\">6<\/a><\/p>\n\n\n\n<p>Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kamyar&nbsp;Seyed Ghasemipour, Burcu&nbsp;Karagol Ayan, S.&nbsp;Sara Mahdavi, Rapha&nbsp;Gontijo Lopes, Tim Salimans, Jonathan Ho, David&nbsp;J Fleet, and Mohammad Norouzi. Photorealistic text-to-image diffusion models with deep language understanding. 2022.&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2205.11487\">arXiv:2205.11487<\/a>.<\/p>\n\n\n\n<h2>&nbsp;\u7a33\u5b9a\u6269\u6563\u6a21\u578b\uff08Stable diffusion model\uff09<\/h2>\n\n\n\n<p><br>Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj\u00f6rn Ommer. High-resolution image synthesis with latent diffusion models. 2021.\u00a0<a href=\"https:\/\/arxiv.org\/abs\/2112.10752\">arXiv:2112.10752<\/a>.<\/p>\n\n\n\n<p>DDPM \u6a21\u578b\u5728\u751f\u6210\u56fe\u50cf\u8d28\u91cf\u4e0a\u6548\u679c\u5df2\u7ecf\u975e\u5e38\u597d\uff0c\u4f46\u5b83\u4e5f\u6709\u4e2a\u7f3a\u70b9\uff0c \u90a3\u5c31\u662f\u00a0\u5c3a\u5bf8\u662f\u548c\u56fe\u7247\u4e00\u81f4\u7684\uff0c\u5143\u7d20\u548c\u56fe\u7247\u7684\u50cf\u7d20\u662f\u4e00\u4e00\u5bf9\u5e94\u7684\uff0c \u6240\u4ee5\u79f0 DDPM \u662f\u50cf\u7d20(pixel)\u7a7a\u95f4\u7684\u751f\u6210\u6a21\u578b\u3002 \u6211\u4eec\u77e5\u9053\u4e00\u5f20\u56fe\u7247\u7684\u5c3a\u5bf8\uff0c\u5982\u679c\u60f3\u751f\u6210\u4e00\u5f20\u9ad8\u5c3a\u5bf8\u7684\u56fe\u50cf\uff0c\u00a0\u5f20\u91cf\u5927\u5c0f\u662f\u975e\u5e38\u5927\u7684\uff0c\u8fd9\u5c31\u9700\u8981\u6781\u5927\u7684\u663e\u5361\uff08\u786c\u4ef6\uff09\u8d44\u6e90\uff0c\u5305\u62ec\u8ba1\u7b97\u8d44\u6e90\u548c\u663e\u5b58\u8d44\u6e90\u3002 \u540c\u6837\u7684\uff0c\u5b83\u7684\u8bad\u7ec3\u6210\u672c\u4e5f\u662f\u9ad8\u6602\u7684\u3002\u9ad8\u6602\u7684\u6210\u672c\u6781\u5927\u7684\u9650\u5236\u4e86\u5b83\u5728\u6c11\u7528\u9886\u7528\u7684\u53d1\u5c55<\/p>\n\n\n\n<h2>\u6f5c\u5728\u6269\u6563\u6a21\u578b<\/h2>\n\n\n\n<p>2021\u5e74\u5fb7\u56fd\u6155\u5c3c\u9ed1\u8def\u5fb7\u7ef4\u5e0c-\u9a6c\u514b\u897f\u7c73\u5229\u5b89\u5927\u5b66\u8ba1\u7b97\u673a\u89c6\u89c9\u548c\u5b66\u4e60\u7814\u7a76\u5c0f\u7ec4\uff08\u539f\u6d77\u5fb7\u5821\u5927\u5b66\u8ba1\u7b97\u673a\u89c6\u89c9\u5c0f\u7ec4\uff09\uff0c \u7b80\u79f0 CompVis \u5c0f\u7ec4\uff0c\u53d1\u5e03\u4e86\u8bba\u6587&nbsp;High-Resolution Image Synthesis with Latent Diffusion Models&nbsp;<a href=\"http:\/\/www.zhangzhenhu.com\/aigc\/%E7%A8%B3%E5%AE%9A%E6%89%A9%E6%95%A3%E6%A8%A1%E5%9E%8B.html#footcite-rombach2021highresolution\">1<\/a>\uff0c\u9488\u5bf9\u8fd9\u4e2a\u95ee\u9898\u505a\u4e86\u4e00\u4e9b\u6539\u8fdb\uff0c \u4e3b\u8981\u7684\u6539\u8fdb\u70b9\u6709\uff1a<\/p>\n\n\n\n<ul><li>\u5f15\u5165\u4e00\u4e2a\u81ea\u7f16\u7801\u5668\uff0c\u5148\u5bf9\u539f\u59cb\u5bf9\u8c61\u8fdb\u884c\u538b\u7f29\u7f16\u7801\uff0c\u7f16\u7801\u540e\u7684\u5411\u91cf\u518d\u5e94\u7528\u5230\u6269\u6563\u6a21\u578b\u3002<\/li><li>\u901a\u8fc7\u5728 UNET \u4e2d\u52a0\u5165 Attention \u673a\u5236\uff0c\u5904\u7406\u6761\u4ef6\u53d8\u91cf&nbsp;<\/li><\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u53c2\u8003\uff1a 1\u3001https:\/\/zhuanlan.zhihu.com\/p\/573984443 2\u3001https:\/\/ &hellip; <a href=\"http:\/\/139.9.1.231\/index.php\/2023\/05\/29\/stable-diffusion\/\" class=\"more-link\">\u7ee7\u7eed\u9605\u8bfb<span class=\"screen-reader-text\">stable diffusion\uff1a\u6f5c\u5728\u6269\u6563\u6a21\u578b<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[37,29,4],"tags":[],"_links":{"self":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/15046"}],"collection":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/comments?post=15046"}],"version-history":[{"count":11,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/15046\/revisions"}],"predecessor-version":[{"id":15297,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/15046\/revisions\/15297"}],"wp:attachment":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/media?parent=15046"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/categories?post=15046"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/tags?post=15046"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}