{"id":497,"date":"2024-04-12T00:13:29","date_gmt":"2024-04-11T16:13:29","guid":{"rendered":"http:\/\/blog.cyasylum.top\/?p=497"},"modified":"2024-04-14T18:47:33","modified_gmt":"2024-04-14T10:47:33","slug":"animationgpt-a-aigc-game-fighting-animation-asset-creation-tool-z1kpiqk","status":"publish","type":"post","link":"http:\/\/blog.cyasylum.top\/index.php\/2024\/04\/12\/animationgpt-a-aigc-game-fighting-animation-asset-creation-tool-z1kpiqk\/","title":{"rendered":"AnimationGPT \u2014\u2014 \u4e00\u4e2aAIGC\u6e38\u620f\u52a8\u753b\u8d44\u4ea7\u521b\u5efa\u5de5\u5177"},"content":{"rendered":"<h1>AnimationGPT \u2014\u2014 \u4e00\u4e2aAIGC\u6e38\u620f\u52a8\u753b\u8d44\u4ea7\u521b\u5efa\u5de5\u5177<\/h1>\n<blockquote><p>\u200b<img decoding=\"async\" src=\"http:\/\/kodo.cyasylum.top\/siyuan\/image-20240412000149-vtajrac.png\" alt=\"image\" \/>\u200b<\/p>\n<p>\u987a\u624b\u7ffb\u5230\u4e86\u8fd9\u4e2a\u9879\u76ee\u6700\u521d\u542f\u52a8\u7684\u65f6\u95f4\uff0c\u5df2\u7ecf\u662f\u4e00\u5e74\u56db\u4e2a\u6708\u524d\u4e86\uff0c\u5728AIGC\u5c1a\u672a\u98ce\u9761\u4e4b\u524d\u5f00\u542f\u7684\u63a2\u7d22\uff0c\u5728\u7b2c\u4e00\u8f6eAIGC\u6d6a\u6f6e\u7a0d\u663e\u843d\u5bde\u4e4b\u65f6\u624d\u843d\u5b9a\u3002<\/p>\n<p>\uff08\u59d1\u4e14\u4e5f\u4e0d\u7b97\u843d\u5b9a\uff0c\u8fd8\u6709\u597d\u591a\u771f\u6b63\u503c\u5f97\u63a8\u8fdb\u7684\u63a2\u7d22\u65b9\u5174\u672a\u827e\uff09<\/p>\n<p>\u8fd9\u7bc7\u6587\u7ae0\u662f\u548c\u8d1d\u4f26\u3001\u9999\u9c7c\u540c\u5b66\u4e00\u540c\u5199\u5c31\u7684\uff0c\u4f46\u8fd9\u5e76\u975e\u6700\u7ec8\u7684publish\u7248\u672c\uff0c\u66f4\u504f\u5411\u4e00\u4e2a\u4e2a\u4eba\u7684\u5b58\u6863<\/p>\n<p>\u6700\u7ec8\u7248\u672c\u4f1a\u88ab\u66f4\u65b0\u5230<a href=\"https:\/\/www.zhihu.com\/people\/ni-mo-82-62\" target=\"_blank\"  rel=\"nofollow\" >\u8d1d\u4f26\u7684\u77e5\u4e4e\u8d26\u53f7<\/a>\u548c<a href=\"http:\/\/139.196.16.200\/enUS\/\" target=\"_blank\"  rel=\"nofollow\" >\u9879\u76ee\u7684\u5b98\u65b9\u9875\u9762<\/a>\u4e0a\uff08\u867d\u7136\u73b0\u5728\u57df\u540d\u90fd\u8fd8\u6ca1\u8fc7\u5ba1\uff09<\/p>\n<p>\uff08\u4e24\u5e74\u5165\u884c\u6e38\u620f\u4e4b\u524d\uff0c\u8fd8\u5728\u62dc\u8bfb\u90a3\u7bc7\u7ecf\u5178\u6587\u7ae0\uff0c\u4e24\u5e74\u540e\u5c31\u4e00\u8d77\u5199\u4e1c\u897f\u4e86\uff0c\u8fd8\u633a\u5947\u5999\u7684\uff09<\/p>\n<p>\u200d<\/p>\n<p>\u63a5\u4e0b\u6765\u5e0c\u671b\u8fd8\u6709\u673a\u4f1a\u8fdb\u4e00\u6b65\u63a2\u7d22\uff0c\u5305\u62ec\u4f18\u5316\u6570\u636e\u96c6\uff0c\u901a\u8fc7RWKV\u52a0\u6301\u4f7f\u5176\u83b7\u5f97\u66f4\u590d\u6742\u7684\u57fa\u4e8e\u9700\u6c42\u751f\u6210\u80fd\u529b\uff1b\u4e5f\u5305\u62ec\u6539\u826f\u52a8\u4f5c\u7f16\u7801\u5668\uff0c\u5f15\u5165 SAN \u7b49\u6846\u67b6\u5b9e\u73b0\u52a8\u4f5c\u6a21\u578b\u7684\u5927\u4e00\u7edf\u3002<\/p>\n<p>\u4e0d\u8fc7\u6309\u6211\u8fd9\u4e09\u5929\u6253\u9c7c\u4e24\u5929\u6652\u7f51\u7684\u6027\u5b50\uff0c\u8c01\u77e5\u9053\u5462\uff0c\u603b\u4e4b\u6b22\u8fce\u5927\u4f19\u7d27\u4fcf\u5173\u6ce8\u540e\u7eed\u7684\u6587\u7ae0\u634f\uff01<\/p><\/blockquote>\n<p>\u200d<\/p>\n<h1>\u6458\u8981<\/h1>\n<p>\u968f\u7740\u4eba\u5de5\u667a\u80fd\u9886\u57df\u7684\u8fc5\u901f\u53d1\u5c55\uff0c\u5bf9\u4e8e\u9ad8\u8d28\u91cf\u3001\u98ce\u683c\u5316\u7684\u52a8\u4f5c\u6570\u636e\u96c6\u7684\u9700\u6c42\u65e5\u76ca\u589e\u957f\u3002\u76ee\u524d\uff0c\u5b66\u672f\u754c\u666e\u904d\u4e3b\u8981\u4f7f\u7528\u7684\u53ea\u6709\u4e00\u4e2a\u4e24\u5e74\u524d\u5236\u4f5c\u7684\u5305\u542b 14.6k \u52a8\u4f5c\u7684\u5f00\u6e90\u52a8\u4f5c\u6570\u636e\u96c6 <a href=\"https:\/\/github.com\/EricGuo5513\/HumanML3D\" target=\"_blank\"  rel=\"nofollow\" >HumanML3D<\/a>\u3002\u7531\u4e8e\u52a8\u4f5c\u6570\u91cf\u7a00\u5c11\u3001\u98ce\u683c\u5355\u4e00\u5316\u751f\u6d3b\u5316\uff0c\u8fd9\u6781\u5927\u9650\u5236\u4e86\u7814\u7a76\u7684\u6df1\u5ea6\u4e0e\u5e7f\u5ea6\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0c\u6211\u4eec\u72ec\u7acb\u5236\u4f5c\u5e76\u5f00\u6e90\u4e86\u4e00\u4e2a\u5168\u65b0\u7684\u3001\u7cbe\u5fc3\u6807\u6ce8\u7684\u3001\u6709 14.8k \u52a8\u4f5c\u7684\u98ce\u683c\u5316\u6218\u6597\u52a8\u4f5c\u6570\u636e\u96c6 CombatMotion Dataset(\u4ee5\u4e0b\u7b80\u79f0\"CM \u6570\u636e\u96c6\")\u3002<\/p>\n<p>\u4f20\u7edf\u4e0a\uff0c\u63cf\u8ff0\u4e00\u4e2a\u6218\u6597\u52a8\u4f5c\u662f\u975e\u5e38\u56f0\u96be\u4e14\u6a21\u7cca\u7684\u5de5\u4f5c\uff0c\u8fd9\u5728\u6a21\u578b\u8bad\u7ec3\u548c\u81ea\u52a8\u5316\u751f\u6210\u4e2d\u4f1a\u5bfc\u81f4\u8bed\u4e49\u7406\u89e3\u65b9\u9762\u7684\u4e0d\u4e00\u81f4\u6027\u3002\u4e3a\u4e86\u514b\u670d\u8fd9\u4e00\u6311\u6218\uff0c\u6211\u4eec\u901a\u8fc7\u89c4\u8303\u6807\u51c6\uff0c\u5212\u5206\u51fa\u516b\u5927\u7c7b\u6218\u6597\u52a8\u4f5c\u63cf\u8ff0\u7ef4\u5ea6\uff0c\u5e76\u901a\u8fc7\u9010\u7c7b\u5efa\u7acb\u4e30\u5bcc\u8bcd\u5e93\u7684\u65b9\u5f0f\uff0c\u63d0\u9ad8\u4e86\u6807\u6ce8\u6548\u7387\u548c\u4e00\u81f4\u6027\u3002<\/p>\n<p>\u901a\u8fc7\u5bf9\u8fd1\u671f\u5404\u7c7b\u79d1\u7814\u6210\u679c\u7684\u7efc\u5408\u6bd4\u8f83\u548c\u65b0\u6570\u636e\u96c6\u7684\u5c1d\u8bd5\u6027\u8bad\u7ec3\uff0c\u6211\u4eec\u53d1\u73b0 <a href=\"https:\/\/github.com\/OpenMotionLab\/MotionGPT\" target=\"_blank\"  rel=\"nofollow\" >MotionGPT<\/a> \u6846\u67b6\u5728\u8bed\u4e49\u7406\u89e3\u548c\u4e00\u81f4\u6027\u65b9\u9762\u8868\u73b0\u6700\u4f73\u3002\u57fa\u4e8e MotionGPT \u6846\u67b6\u548c CM \u6570\u636e\u96c6\uff0c\u6211\u4eec\u6210\u529f\u8bad\u7ec3\u51fa\u4e86\u4e00\u4e2a\u80fd\u591f\u751f\u6210\u9ad8\u8d28\u91cf\u6218\u6597\u52a8\u4f5c\u7684\u6587\u672c-\u52a8\u4f5c\u6a21\u578b\uff0c\u5e76\u64b0\u5199\u4e86\u57fa\u4e8e Maya \u7684\u52a8\u4f5c\u91cd\u5b9a\u5411\u811a\u672c\uff0c\u5c06\u5b66\u672f\u754c\u5e38\u7528\u7684 <a href=\"https:\/\/dl.acm.org\/doi\/abs\/10.1145\/3596711.3596800\" target=\"_blank\"  rel=\"nofollow\" >SMPL \u9aa8\u9abc<\/a> \u8f6c\u6362\u4e3a\u5de5\u4e1a\u754c\u5e7f\u6cdb\u4f7f\u7528\u7684 BVH \u683c\u5f0f\u3002<\/p>\n<p>\u7efc\u4e0a\uff0c\u6211\u4eec\u5b9e\u73b0\u4e86\u4e00\u6574\u5957\u53ef\u590d\u5236\u7684\u5de5\u4f5c\u6d41\u89c4\u8303\uff0c\u5305\u62ec\u6570\u636e\u63d0\u53d6\u3001\u6570\u636e\u6807\u6ce8\u3001\u52a8\u4f5c\u91cd\u5b9a\u5411\u3001\u8bad\u7ec3\u3001\u63a8\u7406\u548c\u63a5\u5165\u5546\u4e1a\u5f15\u64ce\u7684\u5168\u8fc7\u7a0b\u3002\u8fd9\u4e00\u7a81\u7834\u4e0d\u4ec5\u4e3a\u8fd0\u52a8\u6570\u636e\u96c6\u7684\u4e30\u5bcc\u5316\u548c\u79d1\u7814\u5de5\u4f5c\u7684\u62d3\u5bbd\u548c\u6df1\u5165\u63d0\u4f9b\u4e86\u575a\u5b9e\u7684\u57fa\u7840\uff0c\u4e5f\u4e3a\u672a\u6765\u7684\u52a8\u4f5c\u751f\u6210\u6280\u672f\u7684\u53d1\u5c55\u548c\u5546\u4e1a\u5e94\u7528\u5f00\u8f9f\u4e86\u65b0\u7684\u9053\u8def\u3002<\/p>\n<p>\u6211\u4eec\u5e0c\u671b\u8fd9\u9879\u5de5\u4f5c\u80fd\u6fc0\u53d1\u66f4\u591a\u7684\u7814\u7a76\u8005\u3001\u5f00\u53d1\u8005\u548c\u4f01\u4e1a\u5171\u540c\u63a2\u7d22\u8fd9\u4e00\u672a\u5145\u5206\u6316\u6398\u7684\u9886\u57df\uff0c\u5171\u540c\u63a8\u52a8\u4eba\u5de5\u667a\u80fd\u548c\u4eba\u7c7b\u52a8\u4f5c\u751f\u6210\u5b9e\u7528\u5316\u7684\u53d1\u5c55\u3002<\/p>\n<h1>\u65e9\u671f\u8c03\u7814<\/h1>\n<p>\u4ece\u73b0\u5728\u5f00\u59cb\u56de\u987e\uff0c2022 \u5e74\u65e0\u7591\u662f\u751f\u6210\u5f0f AI \u5143\u5e74\uff0c\u8fd9\u4e00\u5e74\u95f4\uff0c\u4ece\u56fe\u50cf\u5230\u6587\u672c\u7684\u751f\u6210\uff0c\u4ece Diffusion \u5230 ChatGPT\u3002\u968f\u7740\u4e00\u4f17\u4eae\u773c\u4ea7\u54c1\u7684\u51fa\u73b0\uff0cAI \u4ece\u79d1\u7814\u9886\u57df\u672a\u6765\u53ef\u671f\u7684\u660e\u706f\uff0c\u53d8\u6210\u4e86\u5207\u5b9e\u80fd\u5f71\u54cd\u5de5\u4f5c\u751f\u4ea7\u7684\u91cd\u8981\u5de5\u5177\u3002<\/p>\n<p>\u5bf9\u6e38\u620f\u884c\u4e1a\u800c\u8a00\uff0c\u7b2c\u4e00\u6b21\u6ce8\u610f\u5230 AI \u7684\u80fd\u529b\u8fd8\u662f 2022 \u5e74\u7684 10 \u6708\u3002NovalAI \u7684\u7206\u706b\u548c\u6cc4\u9732\uff0c\u8ba9\u65e9\u5728 8 \u6708\u53d1\u5e03\u7684 Stable Diffusion \u4ece\u4e8c\u6b21\u5143\u7684\u89d2\u5ea6\u771f\u6b63\u5207\u5165\u5927\u4f17\u7684\u89c6\u91ce\uff0c\u800c\u4e0e\u4e4b\u76f8\u5173\u4e00\u4f17\u65b0\u6280\u672f\u7684\u63d0\u51fa\uff0c\u4e5f\u8ba9 Diffusion \u8fd9\u4e00\u67b6\u6784\u6210\u4e3a\u751f\u6210\u5f0f AI \u9886\u57df\u6700\u7099\u624b\u53ef\u70ed\u7684\u8bfe\u9898\u3002<\/p>\n<p>\u4eba\u4eec\u4e0d\u81ea\u89c9\u5730\u8bbe\u60f3\uff0cDiffusion \u9664\u4e86\u7528\u4e8e\u751f\u6210\u56fe\u7247\uff0c\u8fd8\u80fd\u751f\u6210\u54ea\u4e9b\u66f4\u591a\u4e1c\u897f\u3002<\/p>\n<p>\u200d<\/p>\n<p>\u5f53\u7136\uff0c\u4e0a\u8ff0\u8d8b\u52bf\u5728\u5b66\u754c\u66f4\u65e9\u5730\u88ab\u6d1e\u89c1\u3002\u56e0\u6b64\uff0c\u5f53\u6211\u4eec\u5076\u7136\u8bbe\u60f3\u5230\u6211\u4eec\u80fd\u5426\u7528 Diffusion \u7684\u67b6\u6784\u751f\u6210\u8bf8\u5982\u52a8\u4f5c\u3001\u52a8\u7ebf\u7b49\u5e8f\u5217\u5185\u5bb9\u65f6\uff0c\u6211\u4eec\u53d1\u73b0\u4e86\u4e0d\u4e45\u524d\u63d0\u51fa\u7684[<a href=\"https:\/\/arxiv.org\/abs\/2209.14916\" target=\"_blank\"  rel=\"nofollow\" >https:\/\/arxiv.org\/abs\/2209.14916<\/a>]\u4e00\u6587\u3002\u5176\u5f88\u597d\u5730\u7ed3\u5408\u4e86 Diffusion \u7684\u6269\u6563\u751f\u6210\u601d\u8def\u548c Transformer \u7684\u5e8f\u5217\u9884\u6d4b\u80fd\u529b\uff0c\u5c06\u5176\u8fd0\u7528\u5230\u4e86\u4eba\u7c7b\u52a8\u4f5c\u5e8f\u5217\u7684\u751f\u6210\u5f53\u4e2d\u3002\u5448\u73b0\u4e86\u6781\u597d\u7684\u8bed\u8a00\u6307\u5bfc\u52a8\u4f5c\u751f\u6210\u6548\u679c\u3002<\/p>\n<p>\u8fd9\u4f7f\u5f97\u6211\u4eec\u5f00\u59cb\u6784\u60f3\uff0c\u53ef\u4ee5\u9047\u89c1\u8be5\u751f\u6210\u5de5\u4f5c\u6d41\u50cf\u751f\u6210\u56fe\u50cf\u4e00\u822c\u88ab\u5f15\u5165\u6e38\u620f\u751f\u4ea7\u4e2d\u7684\u672a\u6765\u2014\u2014\u76f4\u5230\u6211\u4eec\u53d1\u73b0\u8be5\u9886\u57df\u4e00\u76f4\u6b62\u6b65\u4e8e\u79d1\u7814\u5883\u5730\u7684\u771f\u5b9e\u539f\u56e0\u3002<\/p>\n<p>\u200d<\/p>\n<p>\u6b63\u5982\u4e0d\u6e29\u4e0d\u706b\u7684 GPT-3\uff0c\u5728\u7ecf\u8fc7\u5fae\u8c03\u540e\u5f62\u6210\u5f15\u7206\u751f\u6210\u5f0f AI \u7684 ChatGPT \u4e00\u6837\uff0c\u6211\u4eec\u53ef\u4ee5\u8ba4\u4e3a\uff0cAI \u5b9e\u7528\u5316\u6709\u7740\u4e09\u9a7e\u9a6c\u8f66\uff0c\u540c\u65f6\u4e5f\u662f\u4e09\u5ea7\u5927\u5c71 \u2014\u2014 \u7b97\u529b\u3001\u7b97\u6cd5\u3001\u6570\u636e\u3002<\/p>\n<p>\u5176\u4e2d\u7b97\u529b\u670d\u4ece\u4e8e\u6469\u5c14\u5b9a\u7406\u7684\u5ba2\u89c2\u89c4\u5f8b\uff0c\u7b97\u6cd5\u4f9d\u8d56\u4e00\u4ee3\u4ee3\u79d1\u5b66\u5bb6\u7684\u7814\u7a76\uff0c\u800c\u6570\u636e\u5374\u522b\u65e0\u53ef\u4f9d\u3002\u9664\u4e86\u60f3\u521b\u9020\u65b0\u7684\u9886\u57df\u53e6\u8f9f\u8e4a\u5f84\u53d1\u8bba\u6587\u7684\u4eba\uff0c\u6216\u662f\u7acb\u5fd7\u63a8\u52a8\u884c\u4e1a\u79d1\u7814\u53d1\u5c55\u7684\u5f00\u6e90\u4e3b\u4e49\u8005\u63a8\u51fa\u7684\u6570\u636e\u96c6\uff0c\u6570\u636e\u5f80\u5f80\u662f\u6700\u532e\u4e4f\u3001\u6700\u96be\u53d6\u5f97\u7684\u8d44\u6e90\u3002<\/p>\n<p>\u5982\u56fe\u50cf\u3001\u6587\u672c\u3001\u58f0\u97f3\u6b64\u7c7b\u8d44\u6e90\uff0c\u5c1a\u4e14\u80fd\u4f9d\u636e\u5176\u539f\u751f\u6027\uff0c\u6709\u7740\u7edf\u4e00\u7684\u6570\u636e\u89c4\u8303\uff0c\u76f8\u5bf9\u5bb9\u6613\u91c7\u96c6\u3002\u82e5\u662f\u8fdb\u4e00\u6b65\u8003\u8651\u90a3\u4e9b\u4eba\u4e3a\u89c4\u5b9a\u7684\u4e8c\u7ea7\u6570\u636e\uff0c\u5176\u7b79\u5907\u96be\u5ea6\u53ef\u60f3\u800c\u77e5 \u2014\u2014 \u663e\u7136\uff0c\u52a8\u4f5c\/\u52a8\u753b\u5c31\u662f\u8fd9\u6837\u4e00\u4e2a\u9886\u57df\u3002<\/p>\n<p>\u5f53\u524d\uff0c\u52a8\u4f5c\u751f\u6210\u7684\u56f0\u5883\u5728\u4e8e\uff0c\u8fc7\u4e8e\u532e\u4e4f\u7684\u6570\u636e\uff0c\u4ee5\u53ca\u532e\u4e4f\u6570\u636e\u4e2d\u8fdb\u4e00\u6b65\u5206\u6563\u7684\u8868\u793a\u65b9\u6cd5\u3002<\/p>\n<ul>\n<li>\u5728\u5de5\u4e1a\u754c\uff0c\u52a8\u753b\u5e08\u901a\u8fc7 FBX \u4e43\u81f3\u5176\u5b83\u9ad8\u7ea7\u6587\u4ef6\u50a8\u5b58\u52a8\u753b\u4fe1\u606f\uff0c\u901a\u8fc7\u8bb0\u5f55\u4e0d\u540c\u540d\u5b57\u548c\u8fde\u63a5\u5173\u7cfb\u7684\u9aa8\u9abc\u5176\u65cb\u8f6c\u4e0e\u4f4d\u79fb\uff0c\u5e76\u57fa\u4e8e\u6b64\u4f20\u9012\u52a8\u753b\u6587\u4ef6\u548c\u5408\u4f5c<\/li>\n<li>\u5728\u5de5\u4e1a\u5b66\u672f\u754c\uff08\u4ee5 Siggraph \u4e3a\u9996\u7684\u524d\u6cbf\u56fe\u5f62\u5b66\uff09\uff0c\u4e60\u60ef\u4f7f\u7528 BVH \u7b49\u6570\u636e\u7ed3\u6784\u9ad8\u6548\u50a8\u5b58\u5404\u79cd\u8fd0\u52a8\u4fe1\u606f\u548c\u57fa\u672c\u9aa8\u67b6\uff0c\u4e13\u6ce8\u4e8e\u5bf9\u6570\u636e\u672c\u8eab\u7684\u5b66\u4e60\u548c\u5904\u7406<\/li>\n<li>\u5728\u5b66\u672f\u754c\uff08\u8fd9\u91cc\u6307\u7684\u4e3b\u8981\u662f\u8ba1\u7b97\u673a\u89c6\u89c9\u5b66\u672f\u754c\uff09\uff0c\u5219\u4e50\u4e8e\u57fa\u4e8e SMPL \u7cfb\u7684\u65e2\u6709\u8f6e\u5b50\u548c\u6570\u636e\u8fdb\u884c\u5f00\u53d1\uff0c\u6349\u6478\u7b97\u6cd5\u6a21\u578b\u672c\u8eab\u7684\u6027\u80fd\uff0c\u5e76\u4ee5\u6253\u699c\u53d1\u8bba\u6587\u4e3a\u9996\u8981\u8ffd\u6c42<\/li>\n<\/ul>\n<p>\u800c\u6211\u4eec\u53d1\u73b0\u7684 MDM \u8fd9\u4e00\u5de5\u4f5c\uff0c\u5176\u5219\u662f\u57fa\u4e8e\u89c6\u89c9\u5b66\u754c\u4f7f\u7528\u7684 SMPL \u4f53\u7cfb\uff0c\u800c\u8fd9\u4e00\u4f53\u7cfb\u4e0b\uff0c\u53ef\u7528\u7684\u6570\u636e\u751a\u81f3\u53ea\u6709\u4e00\u5219\u6570\u636e\u96c6 [<a href=\"https:\/\/github.com\/EricGuo5513\/HumanML3D\" target=\"_blank\"  rel=\"nofollow\" >https:\/\/github.com\/EricGuo5513\/HumanML3D<\/a>]<\/p>\n<p>\u8fd9\u662f\u4e00\u5219\u63cf\u8ff0\u4e86\u5404\u8272\u65e5\u5e38\u52a8\u4f5c\u7684\u52a8\u6355\u6570\u636e\u96c6\uff0c\u5176\u56fa\u7136\u5b58\u5728\u4e00\u5b9a\u4e30\u5bcc\u6027\u4ee5\u652f\u6491\u8be5\u9886\u57df\u7684\u79d1\u7814\u3002\u4f46\u4e00\u65e6\u6d89\u53ca\u66f4\u591a\u3001\u7a0d\u6709\u79bb\u7fa4\u7684\u52a8\u4f5c\uff0c\u5176\u5c31\u5df2\u7ecf\u96be\u4ee5\u4ee3\u8868\u751f\u4ea7\u4e2d\u7684\u5b9e\u9645\u9700\u6c42\u4e86\u3002<\/p>\n<p>\u5c3d\u7ba1\u5f53\u524d\u52a8\u4f5c\u6570\u636e\u7684\u57fa\u7840\u8bbe\u65bd\u5c1a\u4e0d\u5b8c\u5584\uff0c\u6211\u4eec\u5bf9\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd\u6280\u672f\u7684\u672a\u6765\u53d1\u5c55\u4ecd\u6301\u4e50\u89c2\u6001\u5ea6\u3002\u4f9d\u636e\u6211\u4eec\u638c\u63e1\u7684\u9ad8\u8d28\u91cf\u6e38\u620f\u9886\u57df\u52a8\u753b\u8d44\u6e90\u548c\u5bf9\u4e8e\u8fd1\u671f\u5404\u7c7b\u7b97\u6cd5\u6846\u67b6\u7684\u5e7f\u6cdb\u7814\u7a76\u3002\u6211\u4eec\u5e0c\u671b\u5373\u4f7f\u5728\u6280\u672f\u7684\u5168\u9762\u6210\u719f\u4e4b\u524d\uff0c\u8fd9\u4e9b\u65e9\u671f\u6210\u679c\u4e5f\u80fd\u4e3a\u8be5\u9886\u57df\u7684\u8fdb\u6b65\u505a\u51fa\u8d21\u732e\u3002<\/p>\n<p>\u4e8e\u662f\uff0c\u6211\u4eec\u5c1d\u8bd5\u5c55\u5f00\u4e86\u5982\u4eca\u7684\u5de5\u4f5c\uff0c\u5e76\u53d6\u5f97\u4e86\u76f8\u5f53\u6709\u8da3\u4e14\u5bcc\u6709\u6210\u6548\u7684\u7ed3\u679c\uff0c\u5c06\u5728\u4e0b\u6587\u4e2d\u8be6\u7ec6\u8ba8\u8bba\u3002<\/p>\n<h1>\u6570\u636e\u96c6\u5236\u4f5c<\/h1>\n<p>\u5728\u6211\u4eec\u7684\u6587\u672c\u751f\u6210\u52a8\u4f5c\u76f8\u5173\u7814\u7a76\u5de5\u4f5c\u4e2d\uff0c\u6570\u636e\u96c6\u7684\u5236\u4f5c\u662f\u4e00\u4e2a\u81f3\u5173\u91cd\u8981\u7684\u73af\u8282\u3002\u76f8\u8f83\u4e8e\u81ea\u7136\u8bed\u8a00\u53ca\u56fe\u7247\u6570\u636e\u5728\u5185\u5bb9\u4e0a\u7684\u4e30\u5bcc\u4e0e\u591a\u6837\uff0c\u5728\u683c\u5f0f\u4e0a\u7684\u6613\u7528\u4e0e\u7edf\u4e00\u3002\u52a8\u4f5c\u6570\u636e\u65e0\u8bba\u662f\u5728\u5b66\u672f\u6291\u6216\u5de5\u4e1a\u9886\u57df\uff0c\u90fd\u5448\u73b0\u51fa\u4e00\u79cd\u6781\u5ea6\u86ee\u8352\u539f\u59cb\u7684\u5f62\u6001\u3002<\/p>\n<p>\u5957\u7528 <a href=\"https:\/\/arxiv.org\/abs\/2212.02837\" target=\"_blank\"  rel=\"nofollow\" >MoFusion<\/a> \u9879\u76ee\u7684\u4e00\u53e5\u8bdd<\/p>\n<blockquote><p>On the other hand, motion datasets are notorious for not following the same convention, due to the specific needs of their creators. In particular, each dataset can use a skeleton with a different number of joints, different joint indices, as well as a different default pose such as the T-pose, the A-pose, or any other poses.<\/p>\n<p>#\u53e6\u4e00\u65b9\u9762\uff0c\u52a8\u4f5c\u6570\u636e\u96c6\u662f\u81ed\u540d\u662d\u8457\u7684\uff0c\u56e0\u4e3a\u521b\u9020\u8005\u7684\u7279\u5b9a\u9700\u6c42\uff0c\u4ed6\u4eec\u5e76\u4e0d\u9075\u5faa\u76f8\u540c\u7684\u7ea6\u5b9a\u3002\u7279\u522b\u662f\uff0c\u6bcf\u4e2a\u6570\u636e\u96c6\u90fd\u53ef\u4ee5\u4f7f\u7528\u5177\u6709\u4e0d\u540c\u5173\u8282\u6570\u91cf\u3001\u4e0d\u540c\u5173\u8282\u7d22\u5f15\u7684\u9aa8\u9abc\uff0c\u4ee5\u53ca\u4e0d\u540c\u7684\u9ed8\u8ba4\u59ff\u52bf\uff0c\u5982 T \u578b\u59ff\u52bf\u3001A \u578b\u59ff\u52bf\u6216\u5176\u4ed6\u4efb\u4f55\u59ff\u52bf\u3002<\/p><\/blockquote>\n<p>\u4e0a\u8ff0\u6bb5\u843d\u4ec5\u63d0\u5230\u4e86\u56e0\u9aa8\u9abc\u62d3\u6251\u7ed3\u6784\u4e0d\u7edf\u4e00\u800c\u5bfc\u81f4\u7684\u7814\u7a76\u56f0\u96be\uff0c\u66f4\u8fdb\u4e00\u6b65\u7684\u96be\u9898\uff0c\u5728\u4e8e\u5f53\u524d\u4e25\u91cd\u7f3a\u4e4f\u7cbe\u5fc3\u6807\u6ce8\u3001\u98ce\u683c\u591a\u6837\u3001\u6570\u91cf\u8db3\u591f\u7684\u52a8\u4f5c\u6570\u636e\u3002<\/p>\n<p>\u5728 2019 \u5e74\uff0c\u7531\u5149\u5b66\u52a8\u6355\u65b9\u6cd5\u5236\u4f5c\u7684\u5305\u542b 11k \u65e5\u5e38\u4eba\u4f53\u6d3b\u52a8\u52a8\u4f5c\u7684 <a href=\"https:\/\/amass.is.tue.mpg.de\/\" target=\"_blank\"  rel=\"nofollow\" >AMASS \u6570\u636e\u96c6<\/a>\u3002\u5728 2022 \u5e74\uff0cEricGuo5513 \u5c06\u8fd9\u4e00\u6570\u636e\u96c6\u7ed3\u5408\u5176\u4ed6\u62d3\u5c55\u4e3a 14.6k \u52a8\u4f5c\uff0c\u5e76\u4e14\u8fdb\u884c\u7c97\u6807\u6ce8\u540e\u5f00\u6e90\uff0c\u8fd9\u5c31\u662f\u5f53\u524d\u5b66\u672f\u754c\u6700\u5e38\u7528\u7684\u52a8\u4f5c\u6570\u636e\u96c6\u2014\u2014 <a href=\"https:\/\/github.com\/EricGuo5513\/HumanML3D\" target=\"_blank\"  rel=\"nofollow\" >HumanML3D<\/a>\u3002<\/p>\n<p>\u4f46\u4e0d\u5e78\u7684\u662f\uff0c\u8fd9\u4e2a\u6570\u636e\u96c6\u7684\u6570\u91cf\u3001\u8d28\u91cf\u3001\u98ce\u683c\u4f9d\u7136\u4e0d\u6ee1\u8db3\u6211\u4eec\"\u9762\u5411\u52a8\u753b\u5e08\"\u7684\u9700\u6c42\u3002\u7b2c\u4e00\u70b9\u6e90\u4e8e\u52a8\u4f5c\u6570\u636e\u672c\u8eab\uff0c\u98ce\u683c\u8fc7\u4e8e\u65e5\u5e38\u5316\uff0c\u4e14\u56e0\u4e3a\u662f\u65e0\u9700\u6c42\u9762\u5411\u7684\u52a8\u6355\u52a8\u4f5c\uff0c\u6240\u4ee5\u4e0d\u4f1a\u51fa\u73b0\u66f4\u81ea\u7531\uff0c\u66f4\u6709\u8868\u73b0\u529b\u4f46\u662f\u4eba\u4f53\u5176\u5b9e\u65e0\u6cd5\u505a\u51fa\u7684\u52a8\u4f5c\uff1b\u7b2c\u4e8c\u70b9\u6e90\u4e8e\u6807\u6ce8\u7684\u4e0d\u51c6\u786e\u6027\u548c\u968f\u610f\u6027\uff0c\u5f53\u6211\u4eec\u9762\u5411\u4eba\u7c7b\u63d0\u51fa\u52a8\u753b\u8bc9\u6c42\u65f6\uff0c\u5e38\u4f1a\u4ee5\u52a8\u4f5c\u7c7b\u578b \u2192 \u624b\u6027 \u2192 \u59ff\u6001 \u2192 \u52a8\u52bf \u2192 \u56db\u80a2\uff0c\u8fd9\u6837\u7531\u7c97\u5230\u7ec6\u7684\u987a\u5e8f\u53bb\u63cf\u8ff0\u4e00\u4e2a\u52a8\u4f5c\uff0c\u8fc7\u4e8e\u65e5\u5e38\u7684\u8bed\u8a00\u98ce\u683c\u4e0d\u5229\u4e8e\u6a21\u578b\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5efa\u7acb\u81ea\u7136\u8bed\u8a00 \u2192 \u52a8\u4f5c\u6620\u5c04\u3002<\/p>\n<p>\u57fa\u4e8e\u4e0a\u8ff0\u4e24\u4e2a\u8ba4\u8bc6\uff0c\u6211\u4eec\u51b3\u5b9a\u4ece\u5934\u5f00\u59cb\uff0c\u72ec\u7acb\u5236\u4f5c\u4e00\u5957\u4ee5\u9ad8\u8d28\u91cf\u6218\u6597\u52a8\u4f5c\u548c\u7b26\u5408\u52a8\u753b\u5e08\u9700\u6c42\u4e60\u60ef\u7cbe\u6807\u6ce8\u7684\u6570\u636e\u96c6\u3002\u5728\u8fd9\u4e2a\u8fc7\u7a0b\u4e2d\u6211\u4eec\u5efa\u7acb\u4e86\u5b8c\u6574\u7684\u6570\u636e\u63d0\u53d6 \u2192 \u91cd\u5b9a\u5411 \u2192 \u89c4\u8303\u6807\u6ce8 \u2192 \u6da6\u8272\u6574\u5408\u7684\u5236\u4f5c\u6846\u67b6\uff0c\u4e3a\u65e5\u540e\u7ee7\u7eed\u62d3\u5c55\u4efb\u610f\u591a\u7684\u65b0\u6570\u636e\u6253\u4e0b\u4e86\u575a\u5b9e\u7684\u57fa\u7840\u3002<\/p>\n<p>\u6211\u4eec\u9996\u5148\u4ece\u9ad8\u8d28\u91cf\u4f5c\u54c1(\u6bd4\u5982 <a href=\"https:\/\/www.mixamo.com\/\" target=\"_blank\"  rel=\"nofollow\" >Maxiamo<\/a>)\u7684\u8d44\u6e90\u4e2d\u63d0\u53d6\u6240\u9700\u7684\u3001\u6218\u6597\u98ce\u683c\u4e3a\u4e3b\u7684\u52a8\u4f5c\u6570\u636e\uff0c\u8fd9\u4e9b\u6570\u636e\u5305\u62ec\u4e86\u5404\u79cd\u52a8\u4f5c\u7684\u7ec6\u8282\u4fe1\u606f\u3002<\/p>\n<p>\u5176\u6b21\uff0c\u4e3a\u4e86\u7edf\u4e00\u52a8\u4f5c\u6570\u636e\u7684\u6807\u51c6\uff0c\u6211\u4eec\u9009\u53d6\u4e86 SMPL(Skinned Multi-Person Linear Model)\u9aa8\u9abc\u4f5c\u4e3a\u57fa\u5e95\uff0c\u8fd9\u4e00\u9aa8\u9abc\u6a21\u578b\u53d1\u5e03\u4e8e 2015 \u5e74\uff0c\u662f\u4e00\u4e2a\u7ebf\u6027\u53ef\u53e0\u52a0\u7684\u901a\u7528\u4eba\u4f53\u8499\u76ae\u9aa8\u9abc\uff0c\u5e26\u6709 10 \u4e2a\u7ef4\u5ea6\u7684\u5f62\u72b6\u53c2\u6570(shape parameters)\u548c 24\u00d73 \u7ef4\u5ea6\u7684\u59ff\u6001\u53c2\u6570(pose parameters)\u3002\u8bbe\u5b9a\u7684\u57fa\u6a21\u7248\u662f\u901a\u8fc7\u7edf\u8ba1\u5927\u91cf\u7684\u771f\u5b9e\u4eba\u4f53\uff0c\u5236\u4f5c\u6a21\u578b\u540e\u53d6\u5e73\u5747\u503c\u5f97\u5230\u7684\u5f62\u72b6\u3002\u901a\u8fc7\u5bf9\u4e3b\u8981\u5f62\u72b6\u6210\u5206(Principal Shape Components)\u6216\u8005\u79f0\u4e4b\u4e3a\u7aef\u70b9\u504f\u79fb(Vertex Deviations)\u8fdb\u884c\u7ebf\u6027\u7ec4\u5408\uff0c\u5e76\u4e14\u5728\u57fa\u6a21\u7248\u4e0a\u8fdb\u884c\u53e0\u52a0\uff0c\u6211\u4eec\u5c31\u5f62\u6210\u4e86\u9759\u9ed8\u59ff\u6001\u7684\u6a21\u578b\u3002\u662f\u76ee\u524d\u6700\u5e38\u7528\u4e8e\u5b66\u672f\u79d1\u7814\u754c\u7684\u4eba\u4f53\u9aa8\u9abc\u62d3\u6251\u7ed3\u6784\u3002<\/p>\n<p>\u6211\u4eec\u5229\u7528 MotionBuilder \u5bf9\u63d0\u53d6\u7684\u52a8\u4f5c\u6570\u636e\u8fdb\u884c\u4e86\u9aa8\u9abc\u91cd\u5b9a\u5411\uff0c\u5c06\u5176\u5168\u90e8\u91cd\u5b9a\u5411\u81f3 SMPL \u9aa8\u9abc\u6a21\u578b\uff0c\u8fd9\u4e00\u6b65\u9aa4\u786e\u4fdd\u4e86\u6570\u636e\u7684\u4e00\u81f4\u6027\u548c\u53ef\u590d\u7528\u6027\u3002<\/p>\n<p>\u200b<img decoding=\"async\" src=\"http:\/\/kodo.cyasylum.top\/siyuan\/image-20240411235814-94gedkj.png\" alt=\"image\" \/>\u200b<\/p>\n<p>\u968f\u540e\uff0c\u6211\u4eec\u8fdb\u884c\u6587\u4ef6\u683c\u5f0f\u7684\u8f6c\u6362\uff0c\u5177\u4f53\u662f\u5c06 fbx \u683c\u5f0f\u7684\u52a8\u4f5c\u6570\u636e\u5229\u7528 MayaScript \u8f6c\u6362\u4e3a npy \u683c\u5f0f\uff0c\u4ee5\u4fbf\u4e8e\u540e\u7eed\u5904\u7406\u3002<\/p>\n<p>\u5728\u52a8\u4f5c\u6570\u636e\u7684\u6807\u6ce8\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u7ec6\u5316\u4e86\u603b\u5171\u516b\u4e2a\u5b57\u6bb5\u5206\u7c7b\uff0c\u5305\u62ec\u52a8\u4f5c\u7c7b\u578b\u3001\u6b66\u5668\u7c7b\u578b\u3001\u653b\u51fb\u7c7b\u578b\u3001\u65b9\u4f4d\u7c7b\u578b\u3001\u529b\u91cf\u611f\u5f62\u5bb9\u8bcd\u3001\u901f\u5ea6\u611f\u5f62\u5bb9\u8bcd\u3001\u6a21\u7cca\u5f62\u5bb9\u8bcd\u4ee5\u53ca\u59ff\u52bf\u5f62\u5bb9\u53e5\u3002\u8fd9\u4e9b\u5206\u7c7b\u7ef4\u5ea6\u4e0d\u4ec5\u6db5\u76d6\u4e86\u52a8\u4f5c\u6570\u636e\u7684\u5404\u4e2a\u65b9\u9762\uff0c\u4e5f\u4e3a\u540e\u7eed\u7684\u6570\u636e\u5206\u6790\u548c\u6a21\u578b\u8bad\u7ec3\u63d0\u4f9b\u4e86\u4e30\u5bcc\u7684\u6807\u6ce8\u4fe1\u606f\uff0c\u5236\u5b9a\u4e86\u5b8c\u6574\u7684\u6807\u6ce8\u89c4\u8303\u3002\u4e0b\u9762\u5c06\u4ecb\u7ecd\u6bcf\u4e2a\u5b57\u6bb5\u5bf9\u5e94\u7684\u8868\u5f81\u6a21\u5f0f\uff1a<\/p>\n<ul>\n<li>\u52a8\u4f5c\u7c7b\u578b\u2014\u2014\u5bf9\u4e8e\u4efb\u610f\u7c7b\u578b\u52a8\u4f5c\u6709\u6548\uff0c\u8868\u5f81\u52a8\u4f5c\u6a21\u5f0f\u7684\u5927\u7c7b\u5f52\u5c5e<\/li>\n<li>\u6b66\u5668\u7c7b\u578b\u2014\u2014\u5bf9\u4e8e\u6218\u6597\u7c7b\u578b\u52a8\u4f5c\u6709\u6548\uff0c\u8868\u5f81\u52a8\u4f5c\u5bf9\u5e94\u7684\u6b66\u5668\u6a21\u7ec4\u7c7b\u578b<\/li>\n<li>\u653b\u51fb\u7c7b\u578b\u2014\u2014\u5bf9\u4e8e\u6218\u6597\u7c7b\u578b\u52a8\u4f5c\u6709\u6548\uff0c\u8868\u5f81\u52a8\u4f5c\u5bf9\u5e94\u7684\u624b\u6027\u53ca\u653b\u51fb\u5f3a\u5ea6\u548c\u653b\u51fb\u8fde\u643a\u6001<\/li>\n<li>\u65b9\u4f4d\u7c7b\u578b\u2014\u2014\u5bf9\u4e8e\u4efb\u610f\u7c7b\u578b\u52a8\u4f5c\u6709\u6548\uff0c\u8868\u5f81\u52a8\u4f5c\u7684\u6839\u9aa8\u9abc\u8fd0\u52a8\u65b9\u5411<\/li>\n<li>\u529b\u91cf\u611f\u5f62\u5bb9\u8bcd\u2014\u2014\u5bf9\u4e8e\u4efb\u610f\u7c7b\u578b\u52a8\u4f5c\u6709\u6548\uff0c\u8868\u5f81\u52a8\u4f5c\u59ff\u6001\u8f7b\u67d4\u6291\u6216\u6c89\u91cd\u7684\u7a0b\u5ea6\u53d8\u5316<\/li>\n<li>\u901f\u5ea6\u611f\u5f62\u5bb9\u8bcd\u2014\u2014\u5bf9\u4e8e\u4efb\u610f\u7c7b\u578b\u52a8\u4f5c\u6709\u6548\uff0c\u8868\u5f81\u52a8\u4f5c\u59ff\u6001\u8fdf\u7f13\u6291\u6216\u8fc5\u6377\u7684\u7a0b\u5ea6\u53d8\u5316<\/li>\n<li>\u6a21\u7cca\u5f62\u5bb9\u8bcd\u2014\u2014\u5bf9\u4e8e\u4efb\u610f\u7c7b\u578b\u52a8\u4f5c\u6709\u6548\uff0c\u8868\u5f81\u52a8\u4f5c\u7684\u5f15\u8d77\u7684\u611f\u53d7\u4e0e\u60c5\u7eea<\/li>\n<li>\u59ff\u52bf\u63cf\u8ff0\u53e5\u2014\u2014\u5bf9\u4e8e\u4efb\u610f\u7c7b\u578b\u52a8\u4f5c\u6709\u6548\uff0c\u8d34\u5408\u751f\u6d3b\u5316\u7684\u81ea\u7136\u8bed\u8a00\uff0c\u8868\u5f81\u52a8\u4f5c\u89e6\u53d1\u540e\u56db\u80a2\u7684\u8fd0\u52a8\u5f62\u6001<\/li>\n<\/ul>\n<p>\u4e3a\u4e86\u66f4\u7cbe\u786e\u5730\u53cd\u6620\u6218\u6597\u7b56\u5212\u7684\u9700\u6c42\uff0c\u6211\u4eec\u6839\u636e\u7b56\u5212\u63d0\u51fa\u7684\u4e60\u60ef\u7528\u8bed\uff0c\u5bf9\u524d 7 \u79cd\u5b57\u6bb5\u5404\u81ea\u5efa\u7acb\u4e86\u4e00\u4e2a\u4e30\u5bcc\u7684\u8bcd\u5e93\uff0c\u4f9d\u7167\u6620\u5c04\u8868\u8fdb\u884c\u4eba\u5de5\u6807\u6ce8\u3002\u4e3a\u4e86\u8fdb\u4e00\u6b65\u63d0\u5347\u6807\u6ce8\u7684\u8d28\u91cf\u548c\u81ea\u7136\u6027\uff0c\u6211\u4eec\u8fd8\u5229\u7528\u4e86 ChatGPT \u8fdb\u884c\u6807\u6ce8\u6da6\u8272\uff0c\u901a\u8fc7\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6280\u672f\u4f18\u5316\u4e86\u6807\u6ce8\u5185\u5bb9\u7684\u8868\u8fbe\u65b9\u5f0f\u3002<\/p>\n<p>\u6807\u6ce8\u8bcd\u5e93\u89c1\u4e0b\u8868\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>Motion Type<\/th>\n<th>Weapon Type<\/th>\n<th>Attack Type<\/th>\n<th>Direction<\/th>\n<th>Power Descriptor<\/th>\n<th>Speed Descriptor<\/th>\n<th>Fuzzy Descriptor<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Idle<\/td>\n<td>Bare Hand<\/td>\n<td>Left-Handed<\/td>\n<td>In-Place<\/td>\n<td>Light-Weighted<\/td>\n<td>Swift<\/td>\n<td>Piercing<\/td>\n<\/tr>\n<tr>\n<td>Get Hit<\/td>\n<td>Sacred Seal<\/td>\n<td>Right-Handed<\/td>\n<td>Towards Left<\/td>\n<td>Steady<\/td>\n<td>Relative Fast<\/td>\n<td>Slash<\/td>\n<\/tr>\n<tr>\n<td>Death<\/td>\n<td>Fist<\/td>\n<td>One-Handed<\/td>\n<td>Towards Right<\/td>\n<td>Heavy-Weighted<\/td>\n<td>Uniform Speed<\/td>\n<td>Blunt<\/td>\n<\/tr>\n<tr>\n<td>Switch Weapon<\/td>\n<td>Claw<\/td>\n<td>Two-Handed<\/td>\n<td>Forward<\/td>\n<td>Powerless<\/td>\n<td>Relative Slow<\/td>\n<td>Stunned<\/td>\n<\/tr>\n<tr>\n<td>Movement<\/td>\n<td>Great sword\/Colossal sword<\/td>\n<td>Dual Wielding<\/td>\n<td>Backward<\/td>\n<td>Powerful<\/td>\n<td>Slow<\/td>\n<td>Unarmed attack<\/td>\n<\/tr>\n<tr>\n<td>Dodge<\/td>\n<td>Curved Sword\/Curved Greatsword<\/td>\n<td>Light Attack<\/td>\n<td>Left Front<\/td>\n<td>Straightforward<\/td>\n<td>First slow then fast<\/td>\n<td>Featherlike<\/td>\n<\/tr>\n<tr>\n<td>Weapon Attack<\/td>\n<td>Spear<\/td>\n<td>Heavy Attack<\/td>\n<td>Right Front<\/td>\n<td>Charged<\/td>\n<td>First fast then slow<\/td>\n<td>Sluggish<\/td>\n<\/tr>\n<tr>\n<td>Spell<\/td>\n<td>Dagger<\/td>\n<td>Charging<\/td>\n<td>Left Rear<\/td>\n<td>Light Stagger<\/td>\n<td>Cleanly<\/td>\n<td>Uninhibited<\/td>\n<\/tr>\n<tr>\n<td>Use Item<\/td>\n<td>Straight Sword<\/td>\n<td>Charged Heavy Attack<\/td>\n<td>Right Rear<\/td>\n<td>Stagger<\/td>\n<td>Sloppy<\/td>\n<td>Unwilling<\/td>\n<\/tr>\n<tr>\n<td>Interaction<\/td>\n<td>Thrusting Sword<\/td>\n<td>Roll Attack<\/td>\n<td>Mid-Air<\/td>\n<td>Heavy Stagger<\/td>\n<td>Instant Recovery<\/td>\n<td>Smooth and Coherent<\/td>\n<\/tr>\n<tr>\n<td>Be Thrown<\/td>\n<td>Katana<\/td>\n<td>Run Attack<\/td>\n<td>Falling<\/td>\n<td><\/td>\n<td>Recovery<\/td>\n<td>Rigid<\/td>\n<\/tr>\n<tr>\n<td>Social Action<\/td>\n<td>Axe\/Great Axe<\/td>\n<td>Backstep Attack<\/td>\n<td>Be Launched to Air<\/td>\n<td><\/td>\n<td>Slow Recovery<\/td>\n<td>Wide Open<\/td>\n<\/tr>\n<tr>\n<td>Pose<\/td>\n<td>Hammer\/Great Hammer<\/td>\n<td>Junp Attack<\/td>\n<td>Knocked Down<\/td>\n<td><\/td>\n<td><\/td>\n<td>Charging<\/td>\n<\/tr>\n<tr>\n<td>Block\/Guard<\/td>\n<td>Flail<\/td>\n<td>Fall Attack<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Accumulating strength<\/td>\n<\/tr>\n<tr>\n<td>Ride<\/td>\n<td>Whip<\/td>\n<td>Crouch Attack<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Instant Burst<\/td>\n<\/tr>\n<tr>\n<td>Hanging Movement<\/td>\n<td>Staff<\/td>\n<td>Shoot<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Devout<\/td>\n<\/tr>\n<tr>\n<td>Swimming<\/td>\n<td>Torch<\/td>\n<td>Kick<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Crazy<\/td>\n<\/tr>\n<tr>\n<td>Falling<\/td>\n<td>Hand Lantern<\/td>\n<td>Attack Bounce-off<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Dangling<\/td>\n<\/tr>\n<tr>\n<td>Eavesdrop<\/td>\n<td>Ballistae<\/td>\n<td>Deathblow<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Juggling<\/td>\n<\/tr>\n<tr>\n<td>Grappling Hook<\/td>\n<td>Reaper<\/td>\n<td>Holding<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Somersaulted<\/td>\n<\/tr>\n<tr>\n<td>Shinnobi Prosthetic<\/td>\n<td>Halberd<\/td>\n<td>Null<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Dancing<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Bow<\/td>\n<td>Quickstep Attack<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Roundabout<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Shield<\/td>\n<td>Air Attack<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Crawling<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Crossbow<\/td>\n<td>Chasing Slice<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Respect<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Twinblade<\/td>\n<td>Hanging Attack<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Provocation<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Shuriken<\/td>\n<td>Attack underwater<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Angry<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Flame Vent<\/td>\n<td>Stomp<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Cheerful<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Firecracker<\/td>\n<td>Combat Art<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Depressed<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Sabimaru<\/td>\n<td>Transforming Attack<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Gaunt<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Umbrella<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Weak<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Mist Raven<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Painful<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Fan<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Nervous<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Finger Whistle<\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Relaxing<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Gorgeous<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Elegant<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Panicked<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Sudden<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Awkward<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td><\/td>\n<td>Weird<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u63a5\u4e0b\u6765\uff0c\u6211\u4eec\u5bf9\u6807\u6ce8\u8bcd\u6027\u8fdb\u884c\u4e86\u5212\u5206\uff0c\u4ee5\u4fbf\u4e8e\u6a21\u578b\u66f4\u597d\u5730\u7406\u89e3\u548c\u5904\u7406\u6807\u6ce8\u4fe1\u606f\u3002\u57fa\u4e8e\u6b64\uff0c\u6700\u7ec8\u5236\u4f5c\u5f97\u5230\u4e86\u5305\u542b 14.8k \u4e2a\u52a8\u4f5c\u300114.8k \u6761\u6807\u6ce8\u7684\u6218\u6597\u98ce\u683c\u52a8\u4f5c\u6570\u636e\u96c6\u2014\u2014CombatMotionRaw Dataset(CMR Dataset)\uff0c\u76ee\u524d\u8be5\u6570\u636e\u96c6\u5df2\u5f00\u6e90\u5728\u6211\u4eec\u7684 GitHub \u9879\u76ee\u4e2d\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u6211\u4eec\u53cd\u590d\u5c1d\u8bd5\u4e86\u8bad\u7ec3\u96c6\u548c\u6d4b\u8bd5\u96c6\u7684\u4e0d\u540c\u5207\u5272\u65b9\u5f0f\uff0c\u6839\u636e\u8bad\u7ec3\u7ed3\u679c\u7684\u53cd\u9988\u5bf9\u6570\u636e\u96c6\u7684\u5927\u5c0f\u548c\u6807\u6ce8\u5185\u5bb9\u8fdb\u884c\u4e86\u6570\u6b21\u8c03\u4f18\uff0c\u6700\u7ec8\u5f97\u5230\u4e86\u7cbe\u5236\u4f5c\u6570\u636e\u96c6 CombatMotionProcessed Dataset(CMP Dataset)\uff0c\u76ee\u524d\u8be5\u6570\u636e\u96c6\u5df2\u5f00\u6e90\u5728\u6211\u4eec\u7684 GitHub \u9879\u76ee\u4e2d\u3002\u5b83\u62e5\u6709 8.7k \u4e2a\u52a8\u4f5c\u6570\u636e\u548c 26.1k \u6807\u6ce8\u6587\u672c\u3002\u8fd9\u4e2a\u6570\u636e\u96c6\u6700\u7ec8\u786e\u4fdd\u4e86\u6a21\u578b\u8bad\u7ec3\u7684\u6709\u6548\u6027\u548c\u6700\u4f18\u5316\u3002<\/p>\n<p>\u5f53\u7136\u7684\uff0c\u5728\u6570\u636e\u96c6\u5236\u4f5c\u7684\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u4e5f\u9762\u4e34\u4e00\u4e9b\u4e0d\u8db3\u4e4b\u5904\uff0c\u6bd4\u5982\u65e0\u6cd5\u5f97\u5230\u6307\u5b9a\u52a8\u4f5c\u7684\u98ce\u683c\u3001\u9650\u5236\u52a8\u4f5c\u5e27\u6570\u6216\u66ff\u6362\u4e0d\u540c\u7684\u65b9\u6848\u7684\u9700\u6c42\u6307\u4ee4\uff0c\u8fd9\u8ba9\u539f\u672c\u9884\u671f\u201c\u4e0e AI \u987a\u7545\u5bf9\u8bdd\u201d\u7684\u76ee\u6807\u96be\u4ee5\u8fbe\u6210\u3002\u4e3a\u4e86\u514b\u670d\u8fd9\u4e9b\u6311\u6218\uff0c\u6211\u4eec\u4e0b\u4e00\u6b65\u8ba1\u5212\u5bfb\u627e\u5b9e\u9645\u843d\u5730\u7684\u52a8\u4f5c\u9700\u6c42\u4f5c\u4e3a\u6807\u6ce8\uff0c\u5e76\u6301\u7eed\u6269\u5927\u6570\u636e\u96c6\u7684\u89c4\u6a21\uff0c\u4ee5\u671f\u8fbe\u5230\u66f4\u9ad8\u7684\u7814\u7a76\u548c\u5e94\u7528\u4ef7\u503c\u3002<\/p>\n<h1>\u6a21\u578b\u9009\u53d6<\/h1>\n<p>\u968f\u7740\u6570\u636e\u96c6\u5236\u5907\u5de5\u4f5c\u7684\u59a5\u5584\u8fdb\u5c55\uff0c\u5728\u65f6\u95f4\u8de8\u5ea6\u4e0a\uff0c\u4e5f\u51fa\u73b0\u4e86\u4e00\u7cfb\u5217\u503c\u5f97\u53c2\u8003\u7684\u65b0\u5de5\u4f5c\uff0c\u4ed6\u4eec\u591a\u6570\u90fd\u88ab\u7eb3\u5165\u4e86\u6211\u4eec\u9009\u62e9\u751f\u6210\u65b9\u6848\u7684\u8003\u91cf\u3002\u5176\u4e2d\u9664 MDM \u7684\u5916\uff0c\u6700\u503c\u5f97\u8ba8\u8bba\u8fd8\u6709 MLD \u548c MotionGPT<\/p>\n<ul>\n<li><a href=\"https:\/\/github.com\/GuyTevet\/motion-diffusion-model\" target=\"_blank\"  rel=\"nofollow\" >GuyTevet\/motion-diffusion-model: The official PyTorch implementation of the paper \"Human Motion Diffusion Model\" (github.com)<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/ChenFengYe\/motion-latent-diffusion\" target=\"_blank\"  rel=\"nofollow\" >ChenFengYe\/motion-latent-diffusion: [CVPR 2023] Executing your Commands via Motion Diffusion in Latent Space\uff0c a fast and high-quality motion diffusion model (github.com)<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/OpenMotionLab\/MotionGPT\" target=\"_blank\"  rel=\"nofollow\" >OpenMotionLab\/MotionGPT: [NeurIPS 2023] MotionGPT: Human Motion as a Foreign Language\uff0c a unified motion-language generation model using LLMs (github.com)<\/a><\/li>\n<\/ul>\n<p>\u8fd9\u4e09\u4efd\u5de5\u4f5c\u5728\u6280\u672f\u601d\u8def\u4e0a\u4e00\u8109\u76f8\u627f\uff0c\u540c\u65f6\u4e5f\u6709\u7740\u5404\u81ea\u7684\u4f18\u7f3a\u70b9\u53ca\u7279\u70b9\u3002<\/p>\n<p>\u7531\u6b64\u51fa\u53d1\uff0c\u6211\u4eec\u5e0c\u671b\u901a\u8fc7\u8ba8\u8bba\u5176\u5f02\u540c\uff0c\u5c55\u671b\u52a8\u4f5c\u751f\u6210\u5de5\u4f5c\u540e\u7eed\u53d1\u5c55\u7684\u53ef\u80fd\u6027\u3002<\/p>\n<p>\u200b<img decoding=\"async\" src=\"http:\/\/kodo.cyasylum.top\/siyuan\/image-20240331221020-twwwy4d.png\" alt=\"image\" \/>\u200b<\/p>\n<p>\u5982\u4e0a\u56fe\u6240\u793a\uff0c\u7b80\u5355\u6982\u62ec\u4e86\u5728 Text2Motion \u8fd9\u4e00\u4efb\u52a1\u4e2d\uff0c\u4e09\u4e2a\u7f51\u7edc\u7684\u67b6\u6784\u6f14\u5316\uff1a<\/p>\n<ul>\n<li>\u5728 MDM \u5230 MLD \u7684\u6f14\u5316\u4e2d\uff0c\u901a\u8fc7\u5f15\u5165 VAE \u964d\u4f4e\u751f\u6210\u6a21\u578b\u7406\u89e3\u548c\u751f\u6210\u52a8\u4f5c\u7684\u95e8\u69db\uff0c\u4ece\u800c\u52a0\u5feb\u751f\u6210\u901f\u5ea6<\/li>\n<li>MLD \u5230 MGPT \u7684\u6f14\u5316\u4e2d\uff0c\u4fdd\u7559\u4e86 VAE \u5bf9\u52a8\u4f5c\u4fe1\u606f\u7684\u84b8\u998f\uff0c\u5e76\u66ff\u6362\u4e86\u5177\u6709\u66f4\u5f3a\u7406\u89e3\u7684\u8bed\u8a00\u6a21\u578b\u751f\u6210<\/li>\n<\/ul>\n<p>\u5f53\u7136\uff0c\u5728\u8fd9\u4e2a\u6f14\u5316\u8fc7\u7a0b\u4e2d\uff0c\u4e5f\u5f15\u5165\u4e86\u4e00\u5b9a\u7684\u7f3a\u70b9\uff0c\u4f8b\u5982 VAE \u7684\u5f15\u5165\u5fc5\u7136\u5bfc\u81f4\u4e86\u538b\u7f29\u4e2d\u7684\u4fe1\u606f\u635f\u5931\uff0c\u57fa\u4e8e\u8bed\u8a00\u6a21\u578b\u7684\u751f\u6210\u76f8\u6bd4 Diffusion \u5fc5\u7136\u964d\u4f4e\u4e86\u968f\u673a\u6027\u3002<\/p>\n<p>\u800c\u5982\u4f55\u8bc4\u4f30\u8fd9\u4e9b\u56e0\u7d20\u5bf9\u6211\u4eec\u8bad\u7ec3\u53ca\u751f\u6210\u6548\u679c\u7684\u5f71\u54cd\uff0c\u65e0\u7591\u662f\u5b9e\u9a8c\u548c\u6280\u672f\u9009\u578b\u4e2d\u6700\u91cd\u8981\u7684\u4e00\u73af\u3002<\/p>\n<h2><strong>\u6a21\u578b\u7684\u6f14\u5316\u4e0e\u9009\u62e9<\/strong><\/h2>\n<p><strong>\u6269\u6563\u6a21\u578b\u4e0e\u751f\u6210\u591a\u6837\u6027<\/strong><\/p>\n<p>\u5bf9 Diffusion \u7684\u629b\u5f03\uff0c\u662f\u4ece\u65e9\u671f\u4f7f\u7528\u6269\u6563\u6a21\u578b\u751f\u6210\u5230\u73b0\u4eca\u4f7f\u7528\u8bed\u8a00\u6a21\u578b\u4e2d\u6700\u5927\u7684\u8f6c\u53d8\u3002<\/p>\n<p>\u6269\u6563\u6a21\u578b\u5728\u56fe\u50cf\u751f\u6210\u9886\u57df\u53d6\u5f97\u4f18\u52bf\uff0c\u662f\u56e0\u4e3a\u5b83\u4eec\u51fa\u8272\u7684\u968f\u673a\u5316\u591a\u6837\u6027\u4ee5\u53ca\u5bf9\u56fe\u50cf\u9ad8\u9891\u7279\u5f81\u7684\u826f\u597d\u5b66\u4e60\u3002\u4f46\u5f53\u8fd9\u4e2a\u4f18\u70b9\u8f6c\u79fb\u5230\u8fd0\u52a8\u751f\u6210\u9886\u57df\u65f6\uff0c\u5c31\u4e0d\u518d\u90a3\u4e48\u7a81\u51fa\u4e86\u3002<\/p>\n<p>\u5bf9\u4e8e\u4eba\u7c7b\u8fd0\u52a8\u800c\u8a00\uff0c\u4e00\u65b9\u9762\u53d7\u9650\u4e8e\u80a2\u4f53\u7269\u7406\u7ea6\u675f\u548c\u52a8\u4f5c\u529f\u80fd\u6027\uff0c\u5f88\u96be\u6709\u8db3\u591f\u4e30\u5bcc\u7684\u52a8\u4f5c\u5f62\u5f0f;\u53e6\u4e00\u65b9\u9762\uff0c\u7531\u4e8e\u7269\u7406\u9650\u5236\uff0c\u8fd0\u52a8\u4e2d\u5f80\u5f80\u7f3a\u4e4f\u9ad8\u9891\u4fe1\u606f\u3002\u8fd9\u4e24\u4e2a\u7279\u70b9\uff0c\u4f7f\u5f97\u6269\u6563\u6a21\u578b\u5728\u8fd0\u52a8\u751f\u6210\u9886\u57df\u7684\u4f18\u52bf\u4e0d\u50cf\u56fe\u50cf\u9886\u57df\u90a3\u4e48\u660e\u663e\u3002<\/p>\n<p><strong>VAE \u4e0e Token \u5316<\/strong><\/p>\n<p>VAE \u6700\u521d\u5728 MLD \u4e2d\u88ab\u5f15\u5165\u4f5c\u4e3a\u4e2d\u95f4\u5c42\uff0c\u7528\u4e8e\u89e3\u51b3\u5728\u590d\u6742 SMPL \u5e8f\u5217\u6570\u636e\u4e0a\u8fd0\u884c\u6269\u6563\u6a21\u578b\u7684\u4f4e\u6548\u95ee\u9898\u3002<\/p>\n<p>VAE \u6ce8\u610f\u5230\u4e86\u4eba\u7c7b\u8fd0\u52a8\u7f3a\u4e4f\u9ad8\u9891\u7279\u5f81\u548c\u8fd0\u52a8\u8303\u56f4\u53d7\u9650\u7684\u95ee\u9898\uff0c\u56e0\u6b64\u5f15\u5165 VAE \u5bf9\u751f\u6210\u76ee\u6807\u8fdb\u884c\u538b\u7f29\u3002\u8fd9\u79cd\u538b\u7f29\u5728 MGPT \u4e2d\u5f97\u5230\u8fdb\u4e00\u6b65\u53d1\u5c55\uff0c\u4f7f\u7528 Token \u5bf9\u8fd0\u52a8\u8fdb\u884c\u66f4\u9ad8\u5c42\u6b21\u7684\u62bd\u8c61\u8868\u5f81\u3002<\/p>\n<p>\u7136\u800c\uff0c\u968f\u7740\u8fd9\u4e00\u8fc7\u7a0b\u7684\u63a8\u8fdb\uff0c\u6211\u4eec\u5bf9\u8fd0\u52a8\u7684\u8868\u8fbe\u8d8a\u53d1\u4f9d\u8d56\u4e8e\u7f16\u7801\u7cfb\u7edf\u7684\u9c81\u68d2\u6027\uff0c\u4e0d\u53ef\u907f\u514d\u5730\u5f15\u5165\u4e86\u4e00\u4e9b\u5931\u771f\u3002\u4f8b\u5982\u5728\u5b9e\u9a8c\u4e2d\uff0cVAE \u8bad\u7ec3\u8d28\u91cf\u5bfc\u81f4\u4e86\u6216\u591a\u6216\u5c11\u7684\u52a8\u4f5c\u6296\u52a8\u3002\u4e43\u81f3\u5f15\u5165 VQVAE \u540e\uff0c\u8fd0\u52a8\u65b9\u5411\u4fe1\u606f\u4e5f\u51fa\u73b0\u4e86\u4e22\u5931\u3002<\/p>\n<p><strong>CLIP \u4e0e\u8de8\u6a21\u6001\u7406\u89e3<\/strong><\/p>\n<p>CLIP \u51fa\u8272\u7684\u8bed\u4e49\u80fd\u529b\u548c DALLE \u7684\u4f18\u5f02\u8868\u73b0\uff0c\u5728\u4e09\u5e74\u524d\u518d\u6b21\u70b9\u71c3\u4e86 CV \u5728\u8de8\u6a21\u6001\u9886\u57df\u7684\u70ed\u60c5\u3002\u4f46\u5728\u67d0\u79cd\u7a0b\u5ea6\u4e0a\uff0c\u5b83\u4e5f\u9650\u5236\u4e86\u76f8\u5173\u9886\u57df\u7684\u7814\u7a76\u8fdb\u5c55\u3002<\/p>\n<p>CLIP \u901a\u8fc7\u5bf9\u6bd4\u5b66\u4e60\u5b9e\u73b0\u4e86\u6587\u672c\u548c\u89c6\u89c9\u6a21\u6001\u7684\u8fde\u63a5\uff0c\u4f46\u7531\u4e8e\u5176\u65e0\u76d1\u7763\u7279\u6027\uff0c\u8fd9\u79cd\u5bf9\u5e94\u662f\u76f8\u5bf9\u7c97\u7cd9\u7684\u3002\u4ee5\u56fe\u50cf\u751f\u6210\u4e3a\u4f8b\uff0c\u751f\u6210\u9ad8\u8d28\u91cf\u56fe\u50cf\u4f9d\u8d56\u4e8e\u5bf9\u56fe\u50cf\u5185\u5bb9\u7684\u8be6\u7ec6\u63cf\u8ff0\u548c\u5bf9\u5173\u952e\u8bcd\u6587\u672c\u7684\u8c03\u6574\u3002\u5b83\u786e\u5b9e\u6709\u63a7\u5236\u9897\u7c92\u5ea6\u66f4\u9ad8\u7684\u4f18\u52bf\uff0c\u4f46\u65e0\u7591\u4e5f\u6548\u7387\u8f83\u4f4e\u3002\u5b83\u7684\u4f18\u52bf\u5728\u4e8e\u9762\u5411\u63cf\u8ff0\u751f\u6210\uff0c\u800c\u96be\u4ee5\u5904\u7406\u66f4\u590d\u6742\u7684\u8bed\u4e49\u9700\u6c42\u3002<\/p>\n<p>\u56e0\u6b64\uff0c\u5f53\u51fa\u73b0\u771f\u6b63\u4f18\u79c0\u7684\u8bed\u8a00\u6a21\u578b\u540e\uff0c\u5bf9\u5b83\u4eec\u8fdb\u884c\u8fed\u4ee3\u662f\u76f8\u5bf9\u4e3b\u6d41\u7684\u601d\u8def\u3002<\/p>\n<p><strong>T5 \u4e0e\u8bed\u4e49\u80fd\u529b\u7684\u63d0\u5347<\/strong><\/p>\n<p>T5 \u7cfb\u5217\u6a21\u578b\u662f\u4e00\u4e2a Encoder-Decoder \u67b6\u6784\u7684\u8bed\u8a00\u6a21\u578b\uff0c\u4e5f\u662f MotionGPT \u4f7f\u7528\u7684\u67b6\u6784\u3002<\/p>\n<p>\u5f97\u76ca\u4e8e\u8bed\u8a00\u6a21\u578b\u7684\u8bed\u4e49\u7406\u89e3\u548c\u751f\u6210\u80fd\u529b\uff0cT5 \u5728 MotionGPT \u4e2d\u540c\u65f6\u53d6\u4ee3\u4e86 MLD \u4e2d\u7684 CLIP \u548c\u6269\u6563\u6a21\u578b\u7684\u6a21\u5757\u3002\u5b83\u5f15\u5165\u4e86\u8fdc\u8d85 CLIP \u7684\u6587\u672c\u7406\u89e3\u80fd\u529b\uff0c\u540c\u65f6\u4e5f\u653e\u5f03\u4e86\u6269\u6563\u6846\u67b6\u591a\u4f59\u7684\u968f\u673a\u5316\u80fd\u529b\u3002<\/p>\n<p>T5 \u6a21\u578b\u7684\u52a0\u5165\u4f7f\u5f97\u57fa\u4e8e\u6587\u672c\u7684\u6307\u4ee4\u5f0f\u3001\u542b\u7f3a\u7701\u7684\u751f\u6210\u6210\u4e3a\u53ef\u80fd\u3002\u8fd9\u4f7f\u6211\u4eec\u80fd\u591f\u7eb3\u5165\u66f4\u591a\u8d34\u8fd1\u5b9e\u9645\u751f\u4ea7\u9700\u6c42\u7684\u751f\u6210\u65b9\u5f0f\u5230\u89c6\u91ce\u4e2d\u3002<\/p>\n<p><strong>ICL \u4e0e\u5bf9\u8bed\u8a00\u6a21\u578b\u80fd\u529b\u7684\u8fdb\u4e00\u6b65\u9700\u6c42<\/strong><\/p>\n<p>\u8bed\u8a00\u6a21\u578b\u7684\u5f15\u5165\u4f7f Prompt \u8bbe\u8ba1\u66f4\u76f4\u89c2\uff0c\u8fd9\u8fdb\u800c\u50ac\u751f\u4e86\u6211\u4eec\u8fdb\u4e00\u6b65\u62d3\u5c55 Prompt \u4ee5\u66f4\u6709\u6548\u63d0\u51fa\u9700\u6c42\u7684\u9700\u8981\u3002\u56f4\u7ed5\u5bf9\u66f4\u591a\u6e38\u620f\u8bbe\u8ba1\u5e08\u4e60\u60ef\u7684\u8c03\u7814\uff0c\u6211\u4eec\u610f\u8bc6\u5230\u8bf8\u5982\u8f93\u5165\u98ce\u683c\u53c2\u8003\u3001\u8f93\u5165\u751f\u6210\u957f\u5ea6\u3001\u57fa\u4e8e\u751f\u6210\u7ed3\u679c\u7684\u5fae\u8c03\u7b49\u4e00\u7cfb\u5217\u9700\u6c42\u3002<\/p>\n<p>\u53ef\u4ee5\u9884\u89c1\u7684\u662f\uff0c\u968f\u7740\u8fd9\u4e9b\u590d\u6742\u9700\u6c42\u7684\u5f15\u5165\uff0c\u7b80\u5355\u7684 Instruct \u5fae\u8c03\u9010\u6e10\u96be\u4ee5\u6db5\u76d6\uff0c\u6211\u4eec\u5c06\u671f\u671b\u751f\u6210\u6a21\u578b\u62e5\u6709\u66f4\u5f3a\u7684\u65f6\u7a7a\u7406\u89e3\u80fd\u529b\u3001\u66f4\u5f3a\u7684\u8fde\u7eed\u5bf9\u8bdd\u4fee\u6539\u80fd\u529b\uff0c\u4ee5\u53ca\u66f4\u5f3a\u7684\u4ece\u4e0a\u4e0b\u6587\u4e2d\u5b66\u4e60\u80fd\u529b\u3002\u8fd9\u53c8\u4f9d\u8d56\u4e8e\u8bed\u8a00\u6a21\u578b\u89c4\u6a21\u7684\u6269\u5927\u3002<\/p>\n<p>\u56e0\u6b64\uff0c\u9009\u7528\u66f4\u5f3a\u5927\u7684\u8bed\u8a00\u6a21\u578b\u6765\u5b9e\u73b0\u751f\u6210\u4efb\u52a1\u6210\u4e3a\u4e86\u76ee\u524d\u503c\u5f97\u63a2\u7d22\u7684\u4e3b\u8981\u65b9\u5411\u4e4b\u4e00\u3002\u65e0\u8bba\u662f\u5c0f\u89c4\u6a21\u5927\u8bcd\u5e93\u6613\u4e8e\u589e\u91cf\u8bad\u7ec3\u7684 RWKV\uff0c\u8fd8\u662f\u5177\u6709\u4e0e Encoder-Decoder \u67b6\u6784\u76f8\u4f3c\u4f18\u52bf\u7684 GLM \u7cfb\u5217\uff0c\u90fd\u662f\u6211\u4eec\u8fdb\u4e00\u6b65\u5c1d\u8bd5\u7684\u9996\u9009\u3002<\/p>\n<h2>MotionGPT: \u57fa\u4e8e\u8fd0\u52a8\u548c\u8bed\u8a00\u7684\u7edf\u4e00\u6846\u67b6<\/h2>\n<p>\u672c\u6587\u914d\u5957\u7684 Demo \u6700\u7ec8\u9009\u7528\u5728 MotionGPT \u6846\u67b6\u4e0b\u8bad\u7ec3\u5e76\u8fd0\u884c\uff0c\u53d6\u5f97\u4e86\u76f8\u5bf9\u7406\u60f3\u7684\u7ed3\u679c\u3002\u6211\u4eec\u5c06\u5728\u8fd9\u4e00\u8282\u7b80\u8981\u4ecb\u7ecdMotionGPT\u7684\u6846\u67b6\u548c\u65b9\u6cd5\u3002<\/p>\n<p>\u200b<img decoding=\"async\" src=\"http:\/\/kodo.cyasylum.top\/siyuan\/image-20240411224017-y9eoduf.png\" alt=\"image\" \/>\u200b<\/p>\n<p>\u5982\u56fe\u6240\u793a\uff0c\u4e3a MotionGPT \u539f\u6587\u732e\u4e2d\u7684\u63d2\u56fe\uff0c\u59a5\u5584\u5c55\u793a\u4e86 MGPT \u7684\u6846\u67b6\u7ed3\u6784\u3002<\/p>\n<p>\u5728MotionGPT\u4e2d\uff0c\u5ef6\u7eed\u4e86MLD\u7684\u601d\u8def\uff0c\u5728VAE\u5bf9\u539f\u59cb\u8fd0\u52a8\u6570\u636e\u538b\u7f29\u7684\u57fa\u7840\u4e0a\u66f4\u8fdb\u4e00\u6b65\uff0c\u5c06VAE\u7684\u7f16\u7801\u7ed3\u679c\u79bb\u6563\u5316\u538b\u7f29\u5230\u7801\u672c\u4e2d\uff0c\u4f7f\u7528VQVAE\u7684\u65b9\u6cd5\u5c06\u8fd0\u52a8\u4fe1\u606fToken\u5316\u3002\u8fd9\u4e00\u601d\u8def\u6e90\u81ea\u4e00\u79cd\u6734\u7d20\u7684\u751f\u6d3b\u76f4\u89c9\u2014\u2014\u6211\u4eec\u80fd\u901a\u8fc7\u80a2\u4f53\u52a8\u4f5c\u4f20\u8fbe\u8bed\u4e49\u4fe1\u606f\uff0c\u90a3\u4e48\u80a2\u4f53\u52a8\u4f5c\u4e5f\u53ef\u4ee5\u88ab\u89c6\u4e3a\u4e00\u79cd\"\u5916\u8bed\"\uff0c\u88ab\u8bed\u8a00\u6a21\u578b\u6240\u7406\u89e3\u548c\u5904\u7406\u3002<\/p>\n<p>MotionGPT\u901a\u8fc7\u5c06\u4e00\u7cfb\u5217\u8fd0\u52a8\u6570\u636e\u8f6c\u6362\u4e3a\u79bb\u6563\u7684\u8fd0\u52a8Token\uff0c\u5e76\u57fa\u4e8e\u8fd9\u4e9bToken\u8fdb\u884c\u8bed\u8a00\u6a21\u578b\u9884\u8bad\u7ec3\uff0c\u4e3a\u8bed\u8a00\u6a21\u578b\u63d0\u4f9b\u4e86\u5bf9\u4eba\u4f53\u52a8\u4f5c\u7406\u89e3\u548c\u751f\u6210\u7684\u80fd\u529b\u3002<\/p>\n<p>\u901a\u8fc7\u8fdb\u4e00\u6b65\u7684\u8bad\u7ec3\u548c\u5fae\u8c03\uff0cMotionGPT\u80fd\u591f\u6709\u6548\u5b9e\u73b0\u57fa\u4e8e\u6307\u4ee4\u7684\u751f\u6210\uff0c\u4e3a\u751f\u6210\u63d0\u4f9b\u4e86\u8d85\u8d8aCLIP\u7684\u7075\u6d3b\u6027\u3002\u4f8b\u5982\uff0c\u5b83\u4e0d\u4ec5\u80fd\u57fa\u4e8e\u8bed\u8a00\u751f\u6210\uff0c\u8fd8\u80fd\u6839\u636e\u9700\u6c42\u586b\u8865\u3001\u5ef6\u7eed\u8fd0\u52a8\u5e8f\u5217\uff0c\u6216\u8005\u6839\u636e\u7ed9\u5b9a\u8fd0\u52a8\u751f\u6210\u76f8\u5e94\u7684\u63cf\u8ff0\u3002<\/p>\n<p>\u8fd9\u4e00\u7279\u6027\u4f7f\u5f97\u751f\u6210\u5f0fAI\u4e3a\u7528\u6237\u63d0\u4f9b\u66f4\u81ea\u7136\u7684\u4ea4\u4e92\u4f53\u9a8c\u6210\u4e3a\u53ef\u80fd\u3002<\/p>\n<h2><strong>\u603b\u7ed3<\/strong><\/h2>\n<p>\u5bf9\u4e8e\u5f53\u524d\u52a8\u4f5c\u751f\u6210\u9886\u57df\u7684\u53d1\u5c55\uff0cMotionGPT \u5f88\u597d\u5730\u5728\u751f\u6210\u591a\u6837\u6027\u4e0e\u6307\u4ee4\u7b26\u5408\u6027\u4e4b\u95f4\u53d6\u5f97\u4e86\u5e73\u8861\u3002\u4f46\u6574\u4f53\u5b66\u672f\u9886\u57df\u5bf9\u5de5\u4e1a\u573a\u666f\u7684\u9700\u6c42\u7406\u89e3\u8f83\u6d45\uff0c\u8fd9\u4e5f\u4e3a\u6211\u4eec\u8fdb\u4e00\u6b65\u63a2\u7d22\u6570\u636e\u96c6\u7684\u8bbe\u8ba1\u548c\u6a21\u578b\u7684\u9009\u53d6\u63d0\u4f9b\u4e86\u4e00\u5b9a\u7684\u7a7a\u95f4\u3002\u8fd9\u4e5f\u5c06\u5bfc\u5411\u6211\u4eec\u63a5\u4e0b\u6765\u7684\u4e00\u7cfb\u5217\u53ef\u80fd\u5de5\u4f5c\u4e0e\u5c55\u671b\u3002<\/p>\n<h1>\u5b9e\u9a8c<\/h1>\n<p>\u5b9e\u9a8c\u73af\u5883\uff1aUbuntu22.04\uff0cNVIDIA GeForce RTX 4090\uff0cCUDA 11.8\uff0c12 vCPU Intel(R) Xeon(R) Platinum 8352V CPU @ 2.10GHz\uff0cPython 3.10\u3002<\/p>\n<p>\u5b9e\u9a8c\u53c2\u6570\uff1a<\/p>\n<blockquote><p><strong>VQVAE<\/strong><\/p>\n<p>batchsize\uff1a256<\/p>\n<p>epoch\uff1a1500<\/p>\n<p>optimizer\uff1aAdamW<\/p>\n<p>learning rate: 2e-4<\/p><\/blockquote>\n<blockquote><p><strong>Pretrain<\/strong><\/p>\n<p>batchsize\uff1a16<\/p>\n<p>epoch\uff1a200<\/p>\n<p>optimizer\uff1aAdamW<\/p>\n<p>learning rate: 2e-4<\/p><\/blockquote>\n<blockquote><p><strong>Finetune<\/strong><\/p>\n<p>batchsize\uff1a16<\/p>\n<p>epoch\uff1a50<\/p>\n<p>optimizer\uff1aAdamW<\/p>\n<p>learning rate: 1e-4<\/p><\/blockquote>\n<p><strong>Evaluation on CMP<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Metric<\/th>\n<th>MotionGPT<\/th>\n<th>MLD<\/th>\n<th>MDM<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Matching Score\u2193<\/td>\n<td>5.426 \u00b1 0.017<\/td>\n<td>5.753 \u00b1 0.019<\/td>\n<td>5.179 \u00b1 0.013<\/td>\n<\/tr>\n<tr>\n<td>Matching Score (Ground Truth)\u2193<\/td>\n<td>5.166 \u00b1 0.012<\/td>\n<td>5.177 \u00b1 0.018<\/td>\n<td>7.220 \u00b1 0.018<\/td>\n<\/tr>\n<tr>\n<td>R_precision (top 1)\u2191<\/td>\n<td>0.044 \u00b1 0.002<\/td>\n<td>0.048 \u00b1 0.002<\/td>\n<td>0.053 \u00b1 0.002<\/td>\n<\/tr>\n<tr>\n<td>R_precision (top 2)\u2191<\/td>\n<td>0.084 \u00b1 0.003<\/td>\n<td>0.089 \u00b1 0.003<\/td>\n<td>0.097 \u00b1 0.003<\/td>\n<\/tr>\n<tr>\n<td>R_precision (top 3)\u2191<\/td>\n<td>0.122 \u00b1 0.003<\/td>\n<td>0.126 \u00b1 0.003<\/td>\n<td>0.136 \u00b1 0.004<\/td>\n<\/tr>\n<tr>\n<td>R_precision (top 1)(Ground Truth)\u2191<\/td>\n<td>0.050 \u00b1 0.002<\/td>\n<td>0.051 \u00b1 0.002<\/td>\n<td>0.030 \u00b1 0.001<\/td>\n<\/tr>\n<tr>\n<td>R_precision (top 2)(Ground Truth)\u2191<\/td>\n<td>0.094 \u00b1 0.002<\/td>\n<td>0.095 \u00b1 0.003<\/td>\n<td>0.063 \u00b1 0.002<\/td>\n<\/tr>\n<tr>\n<td>R_precision (top 3)(Ground Truth)\u2191<\/td>\n<td>0.133 \u00b1 0.003<\/td>\n<td>0.134 \u00b1 0.004<\/td>\n<td>0.096 \u00b1 0.002<\/td>\n<\/tr>\n<tr>\n<td>FID\u2193<\/td>\n<td>0.531 \u00b1 0.018<\/td>\n<td>1.240 \u00b1 0.036<\/td>\n<td>0.019 \u00b1 0.001<\/td>\n<\/tr>\n<tr>\n<td>Diversity\u2192<\/td>\n<td>5.143 \u00b1 0.052<\/td>\n<td>5.269 \u00b1 0.044<\/td>\n<td>5.191 \u00b1 0.036<\/td>\n<\/tr>\n<tr>\n<td>Diversity (Ground Truth)\u2192<\/td>\n<td>5.188 \u00b1 0.070<\/td>\n<td>5.200 \u00b1 0.049<\/td>\n<td>3.364 \u00b1 0.080<\/td>\n<\/tr>\n<tr>\n<td>MultiModality \u2191<\/td>\n<td>1.793 \u00b1 0.094<\/td>\n<td>2.618 \u00b1 0.115<\/td>\n<td>2.463 \u00b1 0.102<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><strong>\u4e3a\u4ec0\u4e48 mGPT\u3001MLD \u548c MDM \u5728 HumanML3D \u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\u5dee\u5f02\u4e0d\u662f\u5f88\u5927\uff0c\u4f46\u5728 CMP \u4e0a\u5dee\u8ddd\u5f88\u5927\uff1f<\/strong><\/p>\n<p>\u4e00\u65b9\u9762\uff0c\u4ece\u6570\u636e\u96c6\u7684\u89d2\u5ea6\u6765\u770b\uff0cHumanML3D \u6570\u636e\u96c6\u5206\u5e03\u6bd4\u8f83\u5747\u5300\uff0c\u5b83\u5305\u542b\u5927\u91cf\u7684\u65e5\u5e38\u52a8\u4f5c\u548c\u4f53\u80b2\u8fd0\u52a8\u52a8\u4f5c\uff0c\u800c\u4e14\u901a\u8fc7\u955c\u50cf\u52a8\u4f5c\u5b9e\u73b0\u4e86\u6570\u636e\u589e\u5f3a\uff0c\u6709\u5c06\u8fd1 29232 \u6761\u6570\u636e\uff0c\u662f\u6211\u4eec\u7684 3.36 \u500d\u3002CMP \u6570\u636e\u96c6\u7684\u5927\u90e8\u5206\u52a8\u4f5c\u6570\u636e\u662f\u5e26\u6709\u6b66\u5668\u7684\u683c\u6597\u52a8\u4f5c\uff0c\u5206\u5e03\u6bd4\u8f83\u96c6\u4e2d\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u4ece\u7b97\u6cd5\u7684\u89d2\u5ea6\u6765\u770b\uff0c\u6211\u4eec\u5728\u9009\u5b9a\u6a21\u578b\u540e\uff0c\u5bf9 MotionGPT \u5728 CMP \u6570\u636e\u96c6\u4e0a\u505a\u4e86\u4f18\u5316\u3002<\/p>\n<p><strong>\u4e3a\u4ec0\u4e48\u4e0d\u5b9e\u73b0\u6570\u636e\u589e\u5f3a\uff1f<\/strong><\/p>\n<p>\u6211\u4eec\u6709\u8003\u8651\u8fc7\u901a\u8fc7 HumanML3D \u4e00\u6837\u7684\u955c\u50cf\u65b9\u6cd5\u5b9e\u73b0\u6570\u636e\u589e\u5f3a\u3002\u4f46\u5728\u901a\u8fc7\u4e00\u6b21\u6b21\u6539\u8fdb\u6807\u6ce8\u3001\u7cbe\u9009\u6570\u636e\u548c\u8bc4\u4f30\u751f\u6210\u6548\u679c\u540e\uff0c\u53d1\u73b0\u6570\u636e\u7684\u8d28\u91cf\u66f4\u91cd\u8981\u3002\u6570\u636e\u589e\u5f3a\u96be\u4ee5\u5b9e\u73b0\u4e3b\u8981\u51fa\u4e8e\u4ee5\u4e0b\u51e0\u70b9\u8003\u8651\uff1a<\/p>\n<ol>\n<li>\u4e00\u4e9b\u52a8\u4f5c\u7684\u7279\u6b8a\u6027\u5bfc\u81f4\u6211\u4eec\u65e0\u6cd5\u6279\u91cf\u5316\u5bf9\u8fd9\u4e2a\u6570\u636e\u96c6\u8fdb\u884c\u955c\u50cf\u5904\u7406\uff0c\u6bd4\u5982\u89d2\u8272\u6b7b\u4ea1\u52a8\u4f5c\uff0c\u5047\u8bbe\u63cf\u8ff0\u8bcd\u662f\u201d\u89d2\u8272\u8eab\u4f53\u5411\u4e0b\u5760\u53bb\uff0c\u762b\u5012\u5728\u5730\u201c\u3002\u5982\u679c\u5bf9\u5176\u955c\u50cf\uff0c\u201d\u5411\u4e0a\u6b7b\u4ea1\u201c\u65e0\u8bba\u662f\u52a8\u4f5c\u8fd8\u662f\u63cf\u8ff0\uff0c\u90fd\u662f\u4e0d\u5408\u7406\u7684\u3002<\/li>\n<li>\u5173\u4e8e\u524d\u540e\u5de6\u53f3\u7684\u5bf9\u79f0\u65b9\u4f4d\u955c\u50cf\u95ee\u9898\uff0c\u6211\u4eec\u5728\u6570\u636e\u96c6\u6807\u6ce8\u4f18\u5316\u4e2d\u589e\u52a0\u4e86\u5bf9 root motion \u7684\u65b9\u4f4d\u6807\u6ce8\uff0c\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\u8fd9\u6837\u7684\u5904\u7406\u65b9\u6cd5\u6709\u5229\u4e8e\u6a21\u578b\u5b66\u4e60\u5230\u6b63\u786e\u7684\u65b9\u4f4d\u4fe1\u606f\u3002<\/li>\n<li>\u4ece\u65f6\u95f4\u53cd\u6f14\u6765\u770b\uff0c\u4e00\u4e9b\u52a8\u4f5c\u901a\u8fc7\u5012\u653e\u53ef\u80fd\u4f1a\u5f71\u54cd\u6a21\u578b\u5bf9\u8bed\u4e49\u7684\u7406\u89e3\uff0c\u6bd4\u5982\u63e1\u4f4f\u957f\u67aa\u51b2\u523a\uff0c\u5012\u653e\u7684\u8bdd\u5c31\u662f\u62ff\u7740\u6b66\u5668\u53cd\u5411\u8dd1\uff0c\u8fd9\u5f88\u5947\u602a\u3002\uff08 **@ \u8d1d\u4f26 ** \u8d1d\u4f6c\u770b\u770b\u53cd\u6f14\u8fd9\u90e8\u5206\u600e\u4e48\u8bf4\u5408\u7406\uff09<\/li>\n<\/ol>\n<p><strong>\u6df7\u5408\u8bad\u7ec3\u5c1d\u8bd5<\/strong><\/p>\n<p>\u4e3a\u4e86\u51cf\u5c11 CMP \u6570\u636e\u96c6\u8fc7\u5c0f\u7684\u95ee\u9898\uff0c\u6211\u4eec\u5c06 HumanML3D\u3001KIT-ML \u548c CMP \u6570\u636e\u96c6\u6df7\u5408\u8d77\u6765\u8bad\u7ec3\u6a21\u578b\uff0c\u5728\u8bc4\u4f30\u6307\u6807\u4e0a\u4f1a\u5e26\u6765\u5de8\u5927\u63d0\u5347\uff0c\u4f46\u8bc4\u4f30\u6307\u6807\u548c\u89c6\u89c9\u6548\u679c\u5e76\u4e0d\u7b49\u4ef7\uff0c\u5bf9\u4e8e\u90e8\u5206\u751f\u6210\u7ed3\u679c\uff0c\u6df7\u5408\u8bad\u7ec3\u7684\u6a21\u578b\u8868\u73b0\u4e0d\u5982\u5355\u72ec\u4f7f\u7528 CMP \u6570\u636e\u96c6\u8bad\u7ec3\u7684\u6a21\u578b\uff0c\u8fd9\u662f\u56e0\u4e3a\u4e0d\u540c\u6570\u636e\u96c6\u52a8\u4f5c\u98ce\u683c\u7684\u5dee\u5f02\u6539\u53d8\u4e86\u6570\u636e\u5206\u5e03\uff0c\u8fdb\u800c\u5f71\u54cd\u4e86\u6a21\u578b\u7684\u6027\u80fd\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u6211\u4eec\u6ce8\u610f\u5230\u4e00\u4e2a\u7279\u522b\u5927\u7684\u6570\u636e\u96c6 <a href=\"https:\/\/github.com\/IDEA-Research\/Motion-X\" target=\"_blank\"  rel=\"nofollow\" >Motion-X<\/a>\uff0c\u8d77\u521d\u50cf\u5c06\u5176\u8f6c\u6362\u6210 HumanML3D \u7684\u683c\u5f0f\uff0c\u7528\u4e8e\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u6216\u8005\u6269\u5145 VQ-VAE \u7684\u7801\u672c\u957f\u5ea6\u6765\u589e\u52a0\u52a8\u4f5c\u7684\u4e30\u5bcc\u6027\u548c\u98ce\u683c\u5316\u7a0b\u5ea6\uff0c\u4f46\u6570\u636e\u683c\u5f0f\u8f6c\u6362\u7684\u5de5\u4f5c\u5931\u8d25\u4e86\u3002<\/p>\n<p><strong>\u5173\u4e8e MLM \u63a9\u7801\u6bd4\u4f8b\u7684\u5c1d\u8bd5<\/strong><\/p>\n<p>MLM\uff08Masked Language Model\uff09\u5728\u8bad\u7ec3\u7684\u8fc7\u7a0b\u4e2d\uff0c\u4f1a\u901a\u8fc7\u968f\u673a\u63a9\u7801\uff08\u901a\u5e38\u63a9\u7801\u6bd4\u4f8b\u8bbe\u7f6e\u4e3a 15%\uff09\u8ba9\u6a21\u578b\u9884\u6d4b\u88ab\u906e\u853d\u90e8\u5206\u7684\u65b9\u5f0f\u8ba9\u6a21\u578b\u5b66\u4e60\u3002\u5982\u679c\u6709\u5927\u91cf\u7684\u6570\u636e\uff0c\u6a21\u578b\u6709\u8db3\u591f\u7684\u4f8b\u5b50\u6765\u5b66\u4e60\u548c\u6cdb\u5316\uff0c\u4f46\u8003\u8651\u5230 CMP \u6570\u636e\u91cf\u8f83\u5c0f\uff0c\u6211\u4eec\u5c1d\u8bd5\u589e\u52a0\u63a9\u7801\u6bd4\u4f8b\u6765\u5e2e\u52a9\u6a21\u578b\u66f4\u597d\u5730\u5b66\u4e60\u6cdb\u5316\u3002\u53d7\u5230<a href=\"https:\/\/arxiv.org\/abs\/2202.08005\" target=\"_blank\"  rel=\"nofollow\" >\u300aShould You Mask 15% in Masked Language Modeling?\u300b<\/a>\u7684\u542f\u53d1\uff0c\u6211\u4eec\u5c06 MLM \u7684\u63a9\u7801\u7387\u63d0\u5347\u5230 40%\uff0c\u4f46\u4ece\u8bc4\u4f30\u6307\u6807\u548c\u89c6\u89c9\u6548\u679c\u4e0a\u770b\u6ca1\u6709\u660e\u663e\u7684\u53d8\u5316\uff0c\u56e0\u6b64\u4ecd\u7136\u91c7\u7528 MotionGPT \u4e2d 15% \u7684\u8bbe\u7f6e\u3002<\/p>\n<p><strong>Text-to-Motion \u4efb\u52a1\u7684\u62c6\u5206<\/strong><\/p>\n<p>\u539f\u7248\u7684 MotionGPT \u4e2d\u5305\u542b\u4e86 Text-to-Motion\u3001Motion-to-Text\u3001Motion Prediction\u3001Motion In-between \u7b49\u8fd0\u52a8\u76f8\u5173\u7684\u4efb\u52a1\uff0c\u800c\u6211\u4eec\u7684\u4efb\u52a1\u53ea\u9700\u8981\u6a21\u578b\u5904\u7406\u597d Text-to-Motion\uff0c\u56e0\u6b64\u5728\u8fd0\u52a8\u8bed\u8a00\u6a21\u578b\u9884\u8bad\u7ec3\u548c\u6307\u4ee4\u5fae\u8c03\u9636\u6bb5\uff0c\u6211\u4eec\u4ece MotionGPT \u6846\u67b6\u4e2d\u5206\u79bb\u51fa text-to-motion \u4efb\u52a1\uff0c\u8fd9\u4f7f\u5f97\u5728\u8ba1\u7b97\u8d44\u6e90\uff08\u4e00\u53f0 RTX4090\uff09\u6709\u9650\u7684\u60c5\u51b5\u4e0b\u52a0\u901f\u4e86\u8bad\u7ec3\u8fc7\u7a0b\uff0c\u539f\u6765 4 \u5929\u7684\u8bad\u7ec3\u8017\u65f6\u51cf\u5c11\u5230 1 \u5929\u4e0d\u5230\uff0c\u800c\u4e14\u8bc4\u4f30\u6307\u6807\u548c\u89c6\u89c9\u6548\u679c\u4e0d\u53d7\u5f71\u54cd\u3002<\/p>\n<h1>\u5c55\u671b<\/h1>\n<p>\u4f9d\u6258\u4e8e MotionGPT \u6846\u67b6\u7684\u5f3a\u5927\u8bed\u4e49\u7406\u89e3\u80fd\u529b\u548c\u7cbe\u5236\u4f5c\u7684 CM \u5927\u578b\u6570\u636e\u96c6\uff0c\u6211\u4eec\u8bad\u7ec3\u51fa\u4e86\u4e00\u4e2a\u9ad8\u8d28\u91cf\u6218\u6597\u52a8\u4f5c\u7684\u6587\u672c-\u52a8\u4f5c\u751f\u6210\u6a21\u578b\u3002\u5c3d\u7ba1\u5982\u6b64\uff0c\u968f\u7740\u6587\u672c\u751f\u6210\u52a8\u4f5c\u6280\u672f\u7684\u8fdb\u4e00\u6b65\u53d1\u5c55\uff0c\u6211\u4eec\u4e5f\u8ba4\u8bc6\u5230\u4e86\u5728\u6a21\u578b\u9002\u5e94\u6027\u548c\u6570\u636e\u96c6\u62d3\u5c55\u6027\u4e0a\u7684\u4e00\u4e9b\u5c40\u9650\u3002\u4e3a\u6b64\uff0c\u6211\u4eec\u8ba1\u5212\u6301\u7eed\u4f18\u5316\u7b97\u6cd5\uff0c\u5e76\u63a2\u7d22\u66f4\u5e7f\u6cdb\u7684\u5e94\u7528\u573a\u666f\u548c\u66f4\u5927\u7684\u5f00\u6e90\u52a8\u4f5c\u6570\u636e\u96c6\u5236\u4f5c\u3002<\/p>\n<p>\u9488\u5bf9\u5de5\u4e1a\u573a\u666f\u7684\u63a5\u5165\uff0c\u6211\u4eec\u610f\u8bc6\u5230\uff0cSMPL \u9aa8\u9abc\u6a21\u578b\u867d\u7136\u4e3a\u62df\u4eba\u52a8\u4f5c\u63d0\u4f9b\u4e86\u5f3a\u5927\u7684\u652f\u6301\uff0c\u4f46\u5176\u5728\u5de5\u4e1a\u4e60\u60ef\u4e0a\u7684\u5c40\u9650\u6027\u4e5f\u9010\u6e10\u51f8\u663e\u3002\u6211\u4eec\u8ba1\u5212\u5c06 SMPL \u6a21\u578b\u66f4\u6df1\u5165\u5730\u4e0e\u6e38\u620f\u5f15\u64ce\u96c6\u6210\uff0c\u4ee5\u5efa\u7acb\u4e00\u4e2a\u5b8c\u6574\u7684\u7528\u6237\u751f\u6210\u6d41\u7a0b\uff0c\u5b9e\u73b0\u4ece\u6587\u672c\u5230\u52a8\u4f5c\u751f\u6210\u7684\u65e0\u7f1d\u8854\u63a5\u3002\u8fd9\u4e00\u6b65\u5c06\u5927\u5e45\u5ea6\u63d0\u5347\u52a8\u4f5c\u751f\u6210\u6280\u672f\u5728\u6e38\u620f\u5f00\u53d1\u3001\u5f71\u89c6\u5236\u4f5c\u7b49\u9886\u57df\u7684\u5b9e\u7528\u6027\u3002<\/p>\n<p>\u5728\u7b97\u6cd5\u4f18\u5316\u65b9\u9762\uff0c\u6211\u4eec\u8ba4\u8bc6\u5230\u5f53\u524d\u4f7f\u7528\u7684\u8bed\u8a00\u6a21\u578b\u96be\u4ee5\u5b8c\u5168\u6ee1\u8db3\u201c\u63d0\u9700\u6c42\u201d\u7684\u5b9e\u9645\u751f\u4ea7\u6807\u51c6\u3002\u56e0\u6b64\uff0c\u6211\u4eec\u8ba1\u5212\u5c1d\u8bd5\u5f15\u5165 <a href=\"https:\/\/wisemodel.cn\/models\/rwkv4fun\/Rwkv-6-world\" target=\"_blank\"  rel=\"nofollow\" >RWKV6 1.5B<\/a>\u7b49\u66f4\u5148\u8fdb\u7684\u8bed\u8a00\u6a21\u578b\uff0c\u501f\u52a9\u5176\u5e7f\u6cdb\u7684\u591a\u8bed\u8a00\u8bcd\u8868\u548c\u76f8\u5bf9\u4f4e\u5ec9\u7684\u8bad\u7ec3\u6210\u672c\uff0c\u8fdb\u4e00\u6b65\u63d0\u9ad8\u751f\u6210\u52a8\u4f5c\u7684\u8d28\u91cf\u548c\u591a\u6837\u6027\u3002\u540c\u65f6\uff0c\u6211\u4eec\u4e5f\u5728\u63a2\u7d22\u662f\u5426\u6709\u66f4\u9002\u5408\u5904\u7406\u6240\u6709\u7c7b\u4eba\u52a8\u4f5c\u7684\u901a\u7528\u8868\u793a\uff0c\u6bd4\u5982\u8003\u8651\u5f15\u5165 <a href=\"https:\/\/arxiv.org\/abs\/2005.05732\" target=\"_blank\"  rel=\"nofollow\" >Skeleton-Aware Networks<\/a> \u67b6\u6784\uff0c\u4f5c\u4e3a <a href=\"https:\/\/arxiv.org\/abs\/1711.00937\" target=\"_blank\"  rel=\"nofollow\" >VQ-VAE<\/a> \u4f7f\u7528\uff0c\u4ee5\u671f\u671b\u66f4\u6709\u6548\u5730\u5b66\u4e60\u548c\u7f16\u89e3\u7801\u9aa8\u9abc\u52a8\u753b\u7684\u7279\u5f81\u3002<\/p>\n<p>\u5bf9\u4e8e\u6570\u636e\u96c6\u8d28\u91cf\u53ca\u5176\u751f\u4ea7\u7ba1\u7ebf\u7684\u4f18\u5316\uff0c\u6211\u4eec\u8ba1\u5212\u8ba9\u6a21\u578b\u6df1\u5165\u611f\u77e5\u9762\u5411\u7b56\u5212\u7684\u8bed\u6599\uff0c\u5f15\u5165\u5e27\u6570\u5fae\u8c03\u3001\u52a8\u753b\u540e\u5904\u7406\uff08\u5982\u52a8\u753b\u7684\u9006\u5411\u52a8\u529b\u5b66 IK \u5904\u7406\uff09\u7b49\u8fdb\u4e00\u6b65\u7684\u6807\u5ea6\u8bbe\u8ba1\u3002\u5bf9\u4e8e\u540c\u4e00\u52a8\u4f5c\u6570\u636e\uff0c\u53ef\u4ee5\u8bbe\u8ba1\u66f4\u591a\u6837\u5316\u7684\u6587\u672c\u6807\u6ce8\uff0c\u4ee5\u63d0\u5347\u6570\u636e\u96c6\u7684\u8d28\u91cf\u548c\u5e94\u7528\u7684\u5e7f\u6cdb\u6027\u3002\u6570\u636e\u6e05\u6d17\u548c\u9884\u5904\u7406\u7684\u5de5\u4f5c\u6d41\u7a0b\u4e5f\u5c06\u8fdb\u4e00\u6b65\u4f18\u5316\uff0c\u4ee5\u786e\u4fdd\u6570\u636e\u7684\u51c6\u786e\u6027\u548c\u6709\u6548\u6027\uff0c\u66f4\u597d\u5730\u6ee1\u8db3\u5b9e\u9645\u751f\u4ea7\u4e2d\u7684\u9700\u6c42\u3002<\/p>\n<p>\u6700\u540e\uff0c\u6211\u4eec\u547c\u5401\u4e1a\u5185\u540c\u4ec1\u5171\u540c\u53c2\u4e0e\u52a8\u4f5c\u6570\u636e\u7684\u5f00\u6e90\u5de5\u4f5c\uff0c\u4ee5\u5171\u540c\u63a8\u52a8\u6587\u672c\u751f\u6210\u52a8\u4f5c\u6280\u672f\u7684\u53d1\u5c55\uff0c\u8ba9\u8fd9\u9879\u6280\u672f\u66f4\u597d\u5730\u670d\u52a1\u4e8e\u66f4\u5e7f\u6cdb\u7684\u5de5\u4e1a\u5e94\u7528\u573a\u666f\uff0c\u63a8\u52a8\u4eba\u5de5\u667a\u80fd\u6280\u672f\u5728\u4eba\u7c7b\u52a8\u4f5c\u751f\u6210\u9886\u57df\u7684\u5b9e\u7528\u5316\u548c\u666e\u53ca\u5316\u3002<\/p>\n<h1>\u9e23\u8c22<\/h1>\n<p>\u5b8c\u6210\u672c\u7bc7\u6587\u7ae0\u7684\u7814\u7a76\u5de5\u4f5c\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u5f97\u5230\u4e86\u8bf8\u591a\u4e13\u4e1a\u4eba\u58eb\u3001\u540c\u4e8b\u3001\u670b\u53cb\u548c\u5e7f\u5927\u6e38\u620f\u7231\u597d\u8005\u7684\u5927\u529b\u652f\u6301\u4e0e\u5e2e\u52a9\u3002\u5728\u6b64\uff0c\u6211\u4eec\u8877\u5fc3\u611f\u8c22\u4ed6\u4eec\u6bcf\u4e00\u4f4d\u7684\u8f9b\u52e4\u5de5\u4f5c\u4e0e\u5b9d\u8d35\u8d21\u732e\u3002<\/p>\n<p>\u611f\u8c22Ol\u00f3rin\u3001\u9057\u5fd8\u7684\u94f6\u7075\u3001\u795e\u6c14\u7684\u732b\u5728\u6570\u636e\u63d0\u53d6\u65b9\u9762\u7684\u5e2e\u52a9\uff0c\u4ed6\u4eec\u7684\u4e13\u4e1a\u80fd\u529b\u5e2e\u52a9\u9879\u76ee\u89e3\u51b3\u4e86\u6700\u68d8\u624b\u7684\u539f\u59cb\u6570\u636e\u96be\u9898\u3002<\/p>\n<p>\u5728\u6570\u636e\u6807\u6ce8\u8fd9\u9879\u8f9b\u82e6\u52b3\u52a8\u8fc7\u7a0b\u4e2d\u505a\u51fa\u6770\u51fa\u8d21\u732e\u7684\u6b8b\u5141\u3001Yuan\u3001\u674f\u5b50\u3001Nori\u3001\u859b\u5b9a\u8c14\u7684\u732b\u3001Elm\u3001\u7ec6\u8282\u3001Chronos\u3001TimEntropy\u3001sPicaLance\u3001\u751c\u5fc3\u3001\u6c34\u679c\u8336\u3001\u5c0f\u6da6\u3001\u67f1\u5b50\u3001\u6c34\u6bcd\u7b49\u4eba\uff0c\u5bf9\u6b64\u6211\u4eec\u8868\u793a\u6700\u8bda\u631a\u7684\u611f\u8c22\u3002<\/p>\n<p>\u963fKita\u5148\u751f\u7684\u4e13\u4e1a\u77e5\u8bc6\u4e3a\u6211\u4eec\u4e30\u5bcc\u6807\u6ce8\u8bcd\u5e93\u63d0\u4f9b\u4e86\u72ec\u5230\u7684\u89c1\u89e3\u4e0e\u521b\u65b0\uff0c\u6211\u4eec\u8868\u793a\u7531\u8877\u7684\u611f\u6fc0\u3002<\/p>\n<p>\u5bf9\u4e8e\u6ce2\u65af\u7334\u5b50\u5148\u751f\u5728\u52a8\u753b\u91cd\u5b9a\u5411\u4e0a\u7684\u5927\u529b\u6280\u672f\u652f\u6301\u548c\u4e0d\u538c\u5176\u70e6\u7684\u53cd\u590d\u4fee\u6b63\u8c03\u4f18\uff0c\u6211\u4eec\u6df1\u8868\u8c22\u610f\u3002<\/p>\n<p>\u611f\u8c22\u80e1\u6210\u540c\u5b66\u5728\u7f51\u9875\u5236\u4f5c\u65b9\u9762\u7684\u7cbe\u6e5b\u6280\u827a\uff0c\u4ee5\u53ca\u65e9\u4e0a\u597d\u5148\u751f\u5728MaxScript\u811a\u672c\u5236\u4f5c\u4e0a\u7684\u5e2e\u52a9\u3002<\/p>\n<p>\u6700\u540e\uff0c\u6211\u4eec\u8fd8\u8981\u7279\u522b\u611f\u8c22Katana\u5148\u751f\u63d0\u4f9b\u7684\u4e13\u4e1a\u6307\u5bfc\u610f\u89c1\uff0c\u4ed6\u7684\u5efa\u8bae\u5bf9\u6211\u4eec\u7684\u7814\u7a76\u8d77\u5230\u4e86\u5173\u952e\u4f5c\u7528\u3002<\/p>\n<p>\u6bcf\u4e00\u4f4d\u7684\u52aa\u529b\u4e0e\u4ed8\u51fa\u90fd\u662f\u6211\u4eec\u7814\u7a76\u6210\u529f\u4e0d\u53ef\u6216\u7f3a\u7684\u4e00\u90e8\u5206\u3002\u6211\u4eec\u5e0c\u671b\u5728\u672a\u6765\u7ee7\u7eed\u5171\u540c\u63a8\u52a8AIGC\u65b9\u5411\u79d1\u7814\u5de5\u4f5c\u7684\u8fdb\u6b65\u3002\u518d\u6b21\u5bf9\u4e0a\u8ff0\u7684\u6bcf\u4e00\u4f4d\u52a9\u529b\u8005\u8868\u793a\u611f\u8c22\uff01<\/p>\n","protected":false},"excerpt":{"rendered":"<p>AnimationGPT \u2014\u2014 \u4e00\u4e2aAIGC\u6e38\u620f\u52a8\u753b\u8d44\u4ea7\u521b\u5efa\u5de5\u5177 \u200b\u200b \u987a\u624b\u7ffb\u5230\u4e86\u8fd9\u4e2a\u9879\u76ee\u6700\u521d\u542f\u52a8\u7684\u65f6\u95f4\uff0c\u5df2\u7ecf\u662f\u4e00\u5e74\u56db\u4e2a\u6708\u524d\u4e86\uff0c\u5728 &#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"emotion":"","emotion_color":"","title_style":"","license":"","footnotes":""},"categories":[1],"tags":[],"class_list":["post-497","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"_links":{"self":[{"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/posts\/497","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/comments?post=497"}],"version-history":[{"count":6,"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/posts\/497\/revisions"}],"predecessor-version":[{"id":504,"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/posts\/497\/revisions\/504"}],"wp:attachment":[{"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/media?parent=497"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/categories?post=497"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/blog.cyasylum.top\/index.php\/wp-json\/wp\/v2\/tags?post=497"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}