{"id":2825,"date":"2025-04-14T15:45:29","date_gmt":"2025-04-14T07:45:29","guid":{"rendered":"https:\/\/www.15zhi.net\/blog\/?p=2825"},"modified":"2025-04-14T15:45:29","modified_gmt":"2025-04-14T07:45:29","slug":"202504-%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-storydiffusion-consistent-self-attention-for-long-range-image-and-video-generation","status":"publish","type":"post","link":"https:\/\/www.15zhi.net\/blog\/202504-%e8%ae%ba%e6%96%87%e7%a0%94%e8%af%bb-storydiffusion-consistent-self-attention-for-long-range-image-and-video-generation\/","title":{"rendered":"202504 \u8bba\u6587\u7814\u8bfb-StoryDiffusion: Consistent Self-Attention for Long-Range Image and Video Generation"},"content":{"rendered":"\n<p id=\"block-cc395e54-1f17-42f8-b92d-cbe08b1e8f64\">\u6765\u6e90\uff1a\u5357\u5f00\u5927\u5b66\u3001\u5b57\u8282\u8df3\u52a8<\/p>\n\n\n\n<p id=\"block-96af701b-1866-4de2-b73a-06d6e20233aa\">\u4f5c\u8005\uff1a<a href=\"https:\/\/arxiv.org\/search\/cs?searchtype=author&amp;query=Zhou,+Y\">\u5468\u7389\u9e4f<\/a>\uff0c\u00a0<a href=\"https:\/\/arxiv.org\/search\/cs?searchtype=author&amp;query=Zhou,+D\">\u5468\u5927\u5168<\/a><\/p>\n\n\n\n<p id=\"block-6ab71e99-a041-452b-80f7-2e10599da252\">\u4e00\u3001\u8bba\u6587\u4e3b\u8981\u5de5\u4f5c<\/p>\n\n\n\n<p id=\"block-b60c42aa-0cf8-49bc-b0f6-684a58fd3f60\">\u8fd9\u7bc7\u6587\u7ae0\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u81ea\u6211\u6ce8\u610f\u529b\u8ba1\u7b97\u65b9\u6cd5\uff0c\u79f0\u4e3a Consistent Self-Attention\uff0c\u5b83\u663e\u7740\u63d0\u9ad8\u4e86\u751f\u6210\u56fe\u50cf\u4e4b\u95f4\u7684\u4e00\u81f4\u6027\uff0c\u5e76\u4ee5\u96f6\u955c\u5934\u7684\u65b9\u5f0f\u589e\u5f3a\u4e86\u6d41\u884c\u7684\u57fa\u4e8e\u9884\u8bad\u7ec3\u7684\u57fa\u4e8e\u6269\u6563\u7684\u6587\u672c\u5230\u56fe\u50cf\u6a21\u578b\u3002\u4e3a\u4e86\u5c06\u65b9\u6cd5\u6269\u5c55\u5230\u8fdc\u7a0b\u89c6\u9891\u751f\u6210\uff0c\u8fdb\u4e00\u6b65\u5f15\u5165\u4e86\u4e00\u4e2a\u65b0\u7684\u8bed\u4e49\u7a7a\u95f4\u65f6\u95f4\u8fd0\u52a8\u9884\u6d4b\u6a21\u5757\uff0c\u540d\u4e3a Semantic Motion Predictor\u3002\u5b83\u7ecf\u8fc7\u8bad\u7ec3\uff0c\u53ef\u4ee5\u4f30\u8ba1\u8bed\u4e49\u7a7a\u95f4\u4e2d\u4e24\u4e2a\u63d0\u4f9b\u7684\u56fe\u50cf\u4e4b\u95f4\u7684\u8fd0\u52a8\u6761\u4ef6\u3002\u8be5\u6a21\u5757\u5c06\u751f\u6210\u7684\u56fe\u50cf\u5e8f\u5217\u8f6c\u6362\u4e3a\u5177\u6709\u5e73\u6ed1\u8fc7\u6e21\u548c\u4e00\u81f4\u4e3b\u9898\u7684\u89c6\u9891\uff0c\u8fd9\u4e9b\u89c6\u9891\u6bd4\u4ec5\u57fa\u4e8e\u6f5c\u5728\u7a7a\u95f4\u7684\u6a21\u5757\u8981\u7a33\u5b9a\u5f97\u591a\uff0c\u5c24\u5176\u662f\u5728\u957f\u89c6\u9891\u751f\u6210\u7684\u4e0a\u4e0b\u6587\u4e2d\u3002\u901a\u8fc7\u5408\u5e76\u8fd9\u4e24\u4e2a\u65b0\u9896\u7684\u7ec4\u4ef6\uff0c\u63d0\u51fa\u4e00\u4e2a\u6846\u67b6\uff08\u79f0\u4e3a StoryDiffusion\uff09\u53ef\u4ee5\u63cf\u8ff0\u4e00\u4e2a\u57fa\u4e8e\u6587\u672c\u7684\u6545\u4e8b\uff0c\u5176\u4e2d\u5305\u542b\u5305\u542b\u4e30\u5bcc\u5185\u5bb9\u7684\u4e00\u81f4\u56fe\u50cf\u6216\u89c6\u9891\u3002<\/p>\n\n\n\n<p id=\"block-486c9636-2792-454e-ab31-1528268cd1e4\">\u4e8c\u3001\u8bba\u6587\u8d21\u732e<\/p>\n\n\n\n<p id=\"block-8662b281-21bf-47a2-97e6-577240bcb9cc\">1.\u4e00\u81f4\u6027\u81ea\u6ce8\u610f\u529b\uff08Consistent Self-Attention\uff09\uff1a\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u81ea\u6ce8\u610f\u529b\u8ba1\u7b97\u65b9\u6cd5\uff0c\u80fd\u591f\u5728\u4e0d\u9700\u8981\u8bad\u7ec3\u7684\u60c5\u51b5\u4e0b\u589e\u5f3a\u751f\u6210\u56fe\u50cf\u4e4b\u95f4\u7684\u4e00\u81f4\u6027\uff0c\u5e76\u4ee5\u96f6\u6837\u672c\u7684\u65b9\u5f0f\u589e\u5f3a\u73b0\u6709\u7684\u9884\u8bad\u7ec3\u6269\u6563\u6a21\u578b\u3002<\/p>\n\n\n\n<p>2.\u8bed\u4e49\u8fd0\u52a8\u9884\u6d4b\u5668\uff08Semantic Motion Predictor\uff09\uff1a\u5f15\u5165\u4e86\u4e00\u4e2a\u65b0\u9896\u7684\u6a21\u5757\uff0c\u7528\u4e8e\u5728\u8bed\u4e49\u7a7a\u95f4\u4e2d\u9884\u6d4b\u4e24\u4e2a\u7ed9\u5b9a\u56fe\u50cf\u4e4b\u95f4\u7684\u8fd0\u52a8\u6761\u4ef6\uff0c\u8be5\u6a21\u5757\u80fd\u591f\u5c06\u751f\u6210\u7684\u56fe\u50cf\u5e8f\u5217\u8f6c\u6362\u6210\u5177\u6709\u5e73\u6ed1\u8fc7\u6e21\u548c\u4e00\u81f4\u4e3b\u4f53\u7684\u89c6\u9891\u3002<\/p>\n\n\n\n<p id=\"block-e7d7305d-7571-4763-a0a8-2ffe8afd605c\">3.StoryDiffusion \u6846\u67b6\uff1a\u7ed3\u5408\u4e0a\u8ff0\u4e24\u4e2a\u65b0\u7ec4\u4ef6\uff0c\u63d0\u51fa\u4e86\u4e00\u4e2a\u6846\u67b6\uff0c\u80fd\u591f\u6839\u636e\u6587\u672c\u6545\u4e8b\u751f\u6210\u5305\u542b\u4e30\u5bcc\u5185\u5bb9\u7684\u4e00\u81f4\u6027\u56fe\u50cf\u6216\u89c6\u9891\u3002<\/p>\n\n\n\n<p id=\"block-b5e8ce28-9cca-4c25-bd22-13626710b4cc\">\u4e09\u3001\u65b9\u6cd5<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/arxiv.org\/html\/2405.01434v1\/x2.png\" alt=\"\u8bf7\u53c2\u9605\u6807\u9898\"\/><\/figure>\n\n\n\n<p id=\"block-1e6a47f5-e6fd-4ade-a5a3-c1d54f3bf9c0\">StoryDiffusion \u7684\u5b9e\u73b0\u65b9\u6cd5\u4e3b\u8981\u4e3a\u4e24\u4e2a\u6838\u5fc3\u7ec4\u4ef6\uff1a<\/p>\n\n\n\n<p id=\"block-1e6a47f5-e6fd-4ade-a5a3-c1d54f3bf9c0\"><strong>\u4e00\u81f4\u6027\u81ea\u5173\u6ce8\uff08Consistent Self-Attention\uff09<\/strong>\uff1a\u8fd9\u662f\u4e00\u79cd\u65b0\u7684\u81ea\u6ce8\u610f\u529b\u8ba1\u7b97\u65b9\u5f0f\uff0c\u80fd\u591f\u5728\u4e0d\u540c\u56fe\u50cf\u95f4\u5efa\u7acb\u8054\u7cfb\uff0c\u4ee5\u589e\u5f3a\u751f\u6210\u56fe\u50cf\u4e2d\u89d2\u8272\u7684\u4e00\u81f4\u6027\u3002\u8fd9\u79cd\u65b9\u6cd5\u4e0d\u9700\u8981\u989d\u5916\u7684\u8bad\u7ec3\u6216\u5fae\u8c03\uff0c\u53ef\u4ee5\u4f5c\u4e3a\u5373\u63d2\u5373\u7528\u7684\u6a21\u5757\u96c6\u6210\u5230\u73b0\u6709\u7684\u56fe\u50cf\u751f\u6210\u6a21\u578b\u4e2d\u3002<\/p>\n\n\n\n<p><strong>\u8bed\u4e49\u8fd0\u52a8\u9884\u6d4b\u5668\uff08Semantic Motion Predictor\uff09<\/strong>\uff1a\u8fd9\u4e2a\u6a21\u5757\u7528\u4e8e\u957f\u8de8\u5ea6\u7684\u89c6\u9891\u751f\u6210\uff0c\u5b83\u901a\u8fc7\u5728\u8bed\u4e49\u7a7a\u95f4\u4e2d\u9884\u6d4b\u4e24\u4e2a\u6761\u4ef6\u56fe\u50cf\u4e4b\u95f4\u7684\u8fd0\u52a8\u6761\u4ef6\uff0c\u5c06\u4e00\u7cfb\u5217\u751f\u6210\u7684\u56fe\u50cf\u8f6c\u6362\u6210\u5177\u6709\u5e73\u6ed1\u8fc7\u6e21\u7684\u89c6\u9891\u3002\u8fd9\u79cd\u65b9\u6cd5\u80fd\u591f\u5728\u4fdd\u6301\u89d2\u8272\u548c\u573a\u666f\u4e00\u81f4\u6027\u7684\u540c\u65f6\uff0c\u751f\u6210\u5177\u6709\u7269\u7406\u610f\u4e49\u548c\u957f\u8ddd\u79bb\u8fd0\u52a8\u7684\u89c6\u9891\u5185\u5bb9\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-6c9d1a033e4a12ae94b840b7dea3d054_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p id=\"block-f0deb681-ae37-4f7e-8e60-6f502579bcd1\">\u56db\u3001\u5b9e\u9a8c<\/p>\n\n\n\n<p id=\"block-6f028f75-9317-4458-8d26-2496b34e3ece\">\u5b9e\u9a8c\u8bc4\u4f30\u4e86StoryDiffusion\u5728\u751f\u6210\u4e3b\u4f53\u4e00\u81f4\u56fe\u50cf\u548c\u8fc7\u6e21\u89c6\u9891\u65b9\u9762\u7684\u6027\u80fd\uff0c\u5e76\u4e0e\u5f53\u524d\u6700\u5148\u8fdb\u7684\u65b9\u6cd5\u8fdb\u884c\u4e86\u6bd4\u8f83\u3002<\/p>\n\n\n\n<p id=\"block-aa85ec6e-f43e-4911-bdb9-e2e94e52fba7\">\u7528\u6237\u7814\u7a76\u663e\u793a\uff0c\u65e0\u8bba\u662f\u5728\u4e3b\u4f53\u4e00\u81f4\u6027\u56fe\u50cf\u751f\u6210\u8fd8\u662f\u8fc7\u6e21\u89c6\u9891\u751f\u6210\u65b9\u9762\uff0cStoryDiffusion\u90fd\u663e\u793a\u51fa\u538b\u5012\u6027\u7684\u4f18\u52bf\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/arxiv.org\/html\/2405.01434v1\/x4.png\" alt=\"\u8bf7\u53c2\u9605\u6807\u9898\"\/><\/figure>\n\n\n\n<p id=\"block-9d1f715b-fbfe-4c0d-a6c2-202a5e774022\">\u4e94\u3001\u603b\u7ed3<\/p>\n\n\n\n<p id=\"block-e7086d6d-7164-43af-ac67-08fab719d74d\">1.\u8fd9\u7bc7\u6587\u7ae0\u63d0\u51fa\u4e86 SynCamMaster \uff0c\u8fd9\u662f\u4e00\u79cd\u65b0\u9896\u7684\u65b9\u6cd5\uff0c\u53ef\u4ee5\u4ee5\u514d\u8bad\u7ec3\u7684\u65b9\u5f0f\u751f\u6210\u4e00\u81f4\u7684\u56fe\u50cf\u6765\u8bb2\u6545\u4e8b\uff0c\u5e76\u5c06\u8fd9\u4e9b\u4e00\u81f4\u7684\u56fe\u50cf\u8f6c\u6362\u4e3a\u89c6\u9891\u3002<\/p>\n\n\n\n<p id=\"block-59a4ac31-2bd5-447c-869a-79ad511cd800\">2.Consistent Self-Attention \u5728\u591a\u4e2a\u56fe\u50cf\u4e4b\u95f4\u5efa\u7acb\u8054\u7cfb\uff0c\u4ee5\u6709\u6548\u5730\u751f\u6210\u5177\u6709\u4e00\u81f4\u9762\u5b54\u548c\u670d\u88c5\u7684\u56fe\u50cf\u3002<\/p>\n\n\n\n<p>3.\u4f7f\u7528\u8bed\u4e49\u8fd0\u52a8\u9884\u6d4b\u5668\u5c06\u8fd9\u4e9b\u56fe\u50cf\u8f6c\u6362\u4e3a\u89c6\u9891\u5e76\u66f4\u597d\u5730\u53d9\u8ff0\u6545\u4e8b.<\/p>\n\n\n\n<p>4.\u5c40\u9650\u6027\uff1a\u5728\u751f\u6210\u4e3b\u4f53\u4e00\u81f4\u56fe\u50cf\u65f6\uff0c\u53ef\u80fd\u5b58\u5728\u4e00\u4e9b\u7ec6\u8282\u4e0a\u7684\u4e0d\u4e00\u81f4\uff0c\u5982\u9886\u5e26\u7b49\uff1b\u5c3d\u7ba1\u53ef\u4ee5\u4f7f\u7528\u6ed1\u52a8\u7a97\u53e3\u65b9\u6cd5\u751f\u6210\u66f4\u957f\u7684\u89c6\u9891\uff0c\u4f46\u8be5\u65b9\u6cd5\u5e76\u975e\u4e13\u95e8\u8bbe\u8ba1\u7528\u4e8e\u957f\u89c6\u9891\u751f\u6210\uff0c\u56e0\u6b64\u5728\u751f\u6210\u975e\u5e38\u957f\u7684\u89c6\u9891\u65f6\u53ef\u80fd\u4e0d\u5b8c\u7f8e\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6765\u6e90\uff1a\u5357\u5f00\u5927\u5b66\u3001\u5b57\u8282\u8df3\u52a8 \u4f5c\u8005\uff1a\u5468\u7389\u9e4f\uff0c\u00a0\u5468\u5927\u5168 \u4e00\u3001\u8bba\u6587\u4e3b\u8981\u5de5\u4f5c \u8fd9\u7bc7\u6587\u7ae0\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u81ea\u6211\u6ce8\u610f\u529b\u8ba1\u7b97\u65b9\u6cd5\uff0c [&hellip;]<\/p>\n","protected":false},"author":4,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-2825","post","type-post","status-publish","format-standard","hentry","category-events"],"_links":{"self":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/2825","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/comments?post=2825"}],"version-history":[{"count":3,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/2825\/revisions"}],"predecessor-version":[{"id":2841,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/2825\/revisions\/2841"}],"wp:attachment":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/media?parent=2825"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/categories?post=2825"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/tags?post=2825"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}