{"id":29757,"date":"2025-04-02T03:04:18","date_gmt":"2025-04-01T19:04:18","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=29757"},"modified":"2025-11-18T17:06:36","modified_gmt":"2025-11-18T09:06:36","slug":"videomind","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/videomind\/","title":{"rendered":"VideoMind"},"content":{"rendered":"

VideoMind \u662f\u4e00\u4e2a\u5f00\u6e90\u7684\u591a\u6a21\u6001AI\u5de5\u5177\uff0c\u4e13\u6ce8\u4e8e\u957f\u89c6\u9891\u7684\u63a8\u7406\u3001\u95ee\u7b54\u548c\u6458\u8981\u751f\u6210\u3002\u5b83\u7531\u9999\u6e2f\u7406\u5de5\u5927\u5b66\u7684\u5218\u6654\uff08Ye Liu\uff09\u548c\u65b0\u52a0\u5761\u56fd\u7acb\u5927\u5b66Show Lab\u7684\u56e2\u961f\u5171\u540c\u5f00\u53d1\u3002\u5de5\u5177\u6a21\u4eff\u4eba\u7c7b\u7406\u89e3\u89c6\u9891\u7684\u65b9\u5f0f\uff0c\u628a\u4efb\u52a1\u62c6\u6210\u89c4\u5212\u3001\u5b9a\u4f4d\u3001\u9a8c\u8bc1\u548c\u56de\u7b54\u7b49\u6b65\u9aa4\uff0c\u9010\u6b65\u89e3\u51b3\u95ee\u9898\u3002\u6838\u5fc3\u6280\u672f\u662fChain-of-LoRA\uff0c\u901a\u8fc7\u4e00\u4e2a\u57fa\u7840\u6a21\u578b\u5207\u6362\u4e0d\u540c\u89d2\u8272\uff0c\u65e2\u9ad8\u6548\u53c8\u8282\u7701\u8d44\u6e90\u3002VideoMind \u57282024\u5e743\u670821\u65e5\u6b63\u5f0f\u53d1\u5e03\u4ee3\u7801\u3001\u6a21\u578b\u548c\u6570\u636e\u96c6\uff0c\u6280\u672f\u62a5\u544a\u4e5f\u4e8e2025\u5e743\u670817\u65e5\u66f4\u65b0\u3002\u7528\u6237\u53ef\u901a\u8fc7GitHub\u6216Hugging Face\u83b7\u53d6\u8d44\u6e90\u3002<\/p>\n

\u8be5\u9879\u76ee\u5e76\u4e0d\u662f\u63d0\u53d6\u97f3\u9891\u8f6c\u6587\u672c\u7406\u89e3\u5185\u5bb9\uff0c\u800c\u662f\u7eaf\u56fe\u5f62\u5316\u4e14\u6309\u7167\u65f6\u95f4\u6233\u51c6\u786e\u7406\u89e3\u89c6\u9891\u5185\u5bb9\uff0c\u53ef\u7528\u4e8e\u89c6\u9891\u81ea\u52a8\u7cbe\u526a\u3001\u6559\u5b66\u573a\u666f\uff0c\u5341\u5206\u5b9e\u7528\uff01<\/p><\/blockquote>\n

\"VideoMind\uff1a\u89c6\u9891\u6309\u65f6\u95f4\u6233\u5b9a\u4f4d\u5185\u5bb9\u4e0e\u95ee\u7b54\u7684\u5f00\u6e90\u9879\u76ee-1\"

\u4f53\u9a8c\u5730\u5740\uff1ahttps:\/\/huggingface.co\/spaces\/yeliudev\/VideoMind-2B<\/p><\/div>\n

 <\/p>\n

\u529f\u80fd\u5217\u8868<\/h2>\n
    \n
  • \u957f\u89c6\u9891\u65f6\u95f4\u5b9a\u4f4d\uff1a\u627e\u5230\u89c6\u9891\u4e2d\u4e0e\u95ee\u9898\u76f8\u5173\u7684\u65f6\u95f4\u6bb5\u3002<\/li>\n
  • \u89c6\u9891\u95ee\u7b54\u529f\u80fd\uff1a\u6839\u636e\u7528\u6237\u63d0\u95ee\uff0c\u4ece\u89c6\u9891\u4e2d\u63d0\u53d6\u51c6\u786e\u7b54\u6848\u3002<\/li>\n
  • \u89c6\u9891\u6458\u8981\u751f\u6210\uff1a\u63d0\u70bc\u89c6\u9891\u5173\u952e\u5185\u5bb9\uff0c\u751f\u6210\u7b80\u6d01\u6982\u8ff0\u3002<\/li>\n
  • \u591a\u6a21\u6001\u4ee3\u7406\u6846\u67b6\uff1a\u5305\u62ec\u89c4\u5212\u8005\u3001\u5b9a\u4f4d\u8005\u3001\u9a8c\u8bc1\u8005\u548c\u56de\u7b54\u8005\uff0c\u534f\u4f5c\u5904\u7406\u4efb\u52a1\u3002<\/li>\n
  • Chain-of-LoRA\u6280\u672f\uff1a\u7528\u5355\u4e00\u6a21\u578b\u5b9e\u73b0\u591a\u529f\u80fd\uff0c\u51cf\u5c11\u8d44\u6e90\u6d88\u8017\u3002<\/li>\n
  • \u652f\u6301\u591a\u79cd\u786c\u4ef6\uff1a\u53ef\u5728NVIDIA GPU\u6216Ascend NPU\u4e0a\u8fd0\u884c\u3002<\/li>\n
  • \u9ad8\u6548\u8bad\u7ec3\u652f\u6301\uff1a\u96c6\u6210DeepSpeed\u3001LoRA\u3001FlashAttention2\u7b49\u6280\u672f\u3002<\/li>\n
  • \u5728\u7ebf\u6f14\u793a\uff1a\u901a\u8fc7Gradio\u754c\u9762\u5feb\u901f\u4f53\u9a8c\u529f\u80fd\u3002<\/li>\n<\/ul>\n

     <\/p>\n

    \u4f7f\u7528\u5e2e\u52a9<\/h2>\n

    VideoMind \u662f\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u7528\u6237\u9700\u8981\u4e0b\u8f7d\u4ee3\u7801\u5e76\u914d\u7f6e\u73af\u5883\u624d\u80fd\u4f7f\u7528\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u64cd\u4f5c\u6307\u5357\uff0c\u6db5\u76d6\u5b89\u88c5\u548c\u529f\u80fd\u4f7f\u7528\u3002<\/p>\n

    \u5b89\u88c5\u6d41\u7a0b<\/h3>\n
      \n
    1. \u514b\u9686\u4ee3\u7801\u4ed3\u5e93<\/strong>
      \n\u5728\u7ec8\u7aef\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\uff0c\u4e0b\u8f7dVideoMind\u6e90\u7801\uff1a<\/li>\n<\/ol>\n
      git clone https:\/\/github.com\/yeliudev\/VideoMind.git\r\n<\/code><\/pre>\n
      \u8fdb\u5165\u9879\u76ee\u76ee\u5f55\uff1a<\/p>\n
      cd VideoMind\r\n<\/code><\/pre>\n
        \n
      1. \u8bbe\u7f6ePython\u73af\u5883<\/strong>
        \n\u4f7f\u7528conda\u521b\u5efa\u865a\u62df\u73af\u5883\uff08\u5efa\u8baePython 3.12\uff09\uff1a<\/li>\n<\/ol>\n
        conda create -n videomind python=3.12 -y\r\nconda activate videomind\r\n<\/code><\/pre>\n
          \n
        1. \u5b89\u88c5\u4f9d\u8d56\u5e93<\/strong>
          \n\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5\u6240\u9700\u4f9d\u8d56\uff1a<\/li>\n<\/ol>\n
          pip install -r requirements.txt\r\n<\/code><\/pre>\n
          \u82e5\u9700\u9ad8\u6548\u8bad\u7ec3\uff0c\u53ef\u5b89\u88c5DeepSpeed\uff1a<\/p>\n
          pip install deepspeed\r\n<\/code><\/pre>\n
            \n
          1. \u4e0b\u8f7d\u6a21\u578b\u548c\u6570\u636e\u96c6<\/strong><\/li>\n<\/ol>\n
              \n
            • \u8bbf\u95eehttps:\/\/huggingface.co\/collections\/yeliudev\/videomind-67dd41f42c57f0e7433afb36\uff0c\u4e0b\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\uff08\u5982VideoMind-2B<\/code>\uff09\uff0c\u89e3\u538b\u5230models<\/code>\u6587\u4ef6\u5939\u3002<\/li>\n
            • \u6570\u636e\u96c6\u53ef\u5728https:\/\/huggingface.co\/datasets\/yeliudev\/VideoMind-Dataset\u4e0b\u8f7d\uff0c\u5305\u62ec27\u4e2a\u89c6\u9891\u6570\u636e\u96c6\u548cVideoMind-SFT\uff08481K\u8bad\u7ec3\u6570\u636e\uff09\uff0c\u653e\u7f6e\u5728data<\/code>\u76ee\u5f55\u3002<\/li>\n<\/ul>\n
                \n
              1. \u6d4b\u8bd5\u5b89\u88c5<\/strong>
                \n\u8fd0\u884c\u6d4b\u8bd5\u811a\u672c\uff0c\u786e\u8ba4\u73af\u5883\u65e0\u8bef\uff1a<\/li>\n<\/ol>\n
                python test.py\r\n<\/code><\/pre>\n
                \u4e3b\u8981\u529f\u80fd\u64cd\u4f5c<\/h3>\n
                1. \u957f\u89c6\u9891\u65f6\u95f4\u5b9a\u4f4d<\/h4>\n
                  \n
                • \u51c6\u5907\u89c6\u9891<\/strong>\uff1a\u5c06\u89c6\u9891\u6587\u4ef6\u653e\u5165data\/videos<\/code>\u76ee\u5f55\uff0c\u652f\u6301MP4\u3001AVI\u7b49\u683c\u5f0f\u3002<\/li>\n
                • \u8fd0\u884c\u547d\u4ee4<\/strong>\uff1a<\/li>\n<\/ul>\n
                  python run_grounding.py --video_path data\/videos\/sample.mp4 --query \"\u627e\u5230\u8ba8\u8bbaAI\u7684\u90e8\u5206\"\r\n<\/code><\/pre>\n
                    \n
                  • \u7ed3\u679c\u8f93\u51fa<\/strong>\uff1a\u8fd4\u56de\u65f6\u95f4\u8303\u56f4\uff0c\u5982\u201c00:05:12 – 00:07:34\u201d\u3002<\/li>\n<\/ul>\n
                    2. \u89c6\u9891\u95ee\u7b54<\/h4>\n
                      \n
                    • \u8f93\u5165\u95ee\u9898<\/strong>\uff1a\u65b0\u5efaquery.txt<\/code>\uff0c\u5199\u5165\u95ee\u9898\uff0c\u4f8b\u5982\u201c\u89c6\u9891\u91cc\u8c01\u662f\u4e3b\u89d2\uff1f\u201d\u3002<\/li>\n
                    • \u6267\u884c\u547d\u4ee4<\/strong>\uff1a<\/li>\n<\/ul>\n
                      python run_qa.py --video_path data\/videos\/sample.mp4 --query_file query.txt\r\n<\/code><\/pre>\n
                        \n
                      • \u83b7\u53d6\u7b54\u6848<\/strong>\uff1a\u7ec8\u7aef\u663e\u793a\u7ed3\u679c\uff0c\u5982\u201c\u4e3b\u89d2\u662f\u674e\u660e\u201d\u3002<\/li>\n<\/ul>\n
                        3. \u89c6\u9891\u6458\u8981\u751f\u6210<\/h4>\n
                          \n
                        • \u8fd0\u884c\u547d\u4ee4<\/strong>\uff1a<\/li>\n<\/ul>\n
                          python run_summary.py --video_path data\/videos\/sample.mp4\r\n<\/code><\/pre>\n
                            \n
                          • \u67e5\u770b\u6458\u8981<\/strong>\uff1a\u7ed3\u679c\u4fdd\u5b58\u4e3asummary.txt<\/code>\uff0c\u5982\u201c\u89c6\u9891\u4ecb\u7ecd\u4e86\u4e00\u573a\u79d1\u6280\u8bb2\u5ea7\u201d\u3002<\/li>\n<\/ul>\n
                            \u7279\u8272\u529f\u80fd\u64cd\u4f5c<\/h3>\n
                            Chain-of-LoRA\u6280\u672f<\/h4>\n
                              \n
                            • VideoMind \u4f7f\u7528\u5355\u4e00\u6a21\u578b\uff0c\u901a\u8fc7LoRA\u9002\u914d\u5668\u5207\u6362\u89d2\u8272\u3002\u914d\u7f6e\u6587\u4ef6\u5728config\/lora.yaml<\/code>\uff0c\u53ef\u8c03\u6574rank<\/code>\u53c2\u6570\u63d0\u5347\u6548\u679c\uff1a<\/li>\n<\/ul>\n
                              lora:\r\nrank: 16\r\n<\/code><\/pre>\n
                              \u5728\u7ebf\u6f14\u793a<\/h4>\n
                                \n
                              • \u65e0\u9700\u672c\u5730\u5b89\u88c5\uff0c\u53ef\u8bbf\u95eehttps:\/\/huggingface.co\/spaces\/yeliudev\/VideoMind-2B\uff0c\u4e0a\u4f20\u89c6\u9891\u5e76\u8f93\u5165\u95ee\u9898\uff0c\u76f4\u63a5\u4f53\u9a8c\u529f\u80fd\u3002\u8be6\u60c5\u89c1docs\/DEMO.md<\/code>\u3002<\/li>\n<\/ul>\n
                                \u8bad\u7ec3\u6a21\u578b<\/h4>\n
                                  \n
                                • \u51c6\u5907\u6570\u636e<\/strong>\uff1a\u4f7f\u7528VideoMind-SFT<\/code>\u6216\u5176\u4ed627\u4e2a\u6570\u636e\u96c6\uff0c\u653e\u5728data<\/code>\u76ee\u5f55\u3002<\/li>\n
                                • \u542f\u52a8\u8bad\u7ec3<\/strong>\uff1a\u8fd0\u884c\uff1a<\/li>\n<\/ul>\n
                                  python train.py --dataset data\/VideoMind-SFT --model models\/VideoMind-2B\r\n<\/code><\/pre>\n
                                    \n
                                  • \u76d1\u63a7\u8bad\u7ec3<\/strong>\uff1a\u652f\u6301Tensorboard\u6216Wandb\u67e5\u770b\u8fdb\u5ea6\uff0c\u5177\u4f53\u914d\u7f6e\u89c1docs\/TRAIN.md<\/code>\u3002<\/li>\n<\/ul>\n
                                    \u6ce8\u610f\u4e8b\u9879<\/h3>\n
                                      \n
                                    • \u89c6\u9891\u6587\u4ef6\u8fc7\u5927\u65f6\uff0c\u5efa\u8bae\u5206\u6bb5\u5904\u7406\uff0c\u907f\u514d\u5185\u5b58\u6ea2\u51fa\u3002<\/li>\n
                                    • Ascend NPU\u7528\u6237\u9700\u5b89\u88c5\u6607\u817e\u9a71\u52a8\uff0c\u8fd0\u884c\u65f6\u52a0\u53c2\u6570--device ascend<\/code>\u3002<\/li>\n
                                    • \u66f4\u591a\u8c03\u8bd5\u4fe1\u606f\u89c1docs\/EVAL.md<\/code>\uff0c\u5305\u62ec\u57fa\u51c6\u6d4b\u8bd5\u6307\u5357\u3002<\/li>\n<\/ul>\n
                                      \u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4\uff0c\u7528\u6237\u53ef\u4ee5\u5feb\u901f\u5b89\u88c5\u5e76\u4f7f\u7528VideoMind\uff0c\u5904\u7406\u957f\u89c6\u9891\u4efb\u52a1\u3002<\/p>\n
                                       <\/p>\n
                                      \u5e94\u7528\u573a\u666f<\/h2>\n
                                        \n
                                      1. \u89c6\u9891\u5185\u5bb9\u5206\u6790<\/strong>
                                        \n\u7528\u6237\u53ef\u4ee5\u7528VideoMind\u5206\u6790YouTube\u957f\u89c6\u9891\uff0c\u5b9a\u4f4d\u5173\u952e\u7247\u6bb5\u6216\u751f\u6210\u6458\u8981\uff0c\u8282\u7701\u89c2\u770b\u65f6\u95f4\u3002<\/li>\n
                                      2. \u6559\u80b2\u4e0e\u7814\u7a76<\/strong>
                                        \n\u5b66\u751f\u6216\u7814\u7a76\u8005\u5904\u7406\u8bb2\u5ea7\u89c6\u9891\u65f6\uff0c\u53ef\u63d0\u95ee\u201c\u8bb2\u4e86\u54ea\u4e9b\u7b97\u6cd5\uff1f\u201d\u5e76\u83b7\u53d6\u7b54\u6848\uff0c\u6574\u7406\u7b14\u8bb0\u66f4\u9ad8\u6548\u3002<\/li>\n
                                      3. \u5185\u5bb9\u5ba1\u6838<\/strong>
                                        \n\u5e73\u53f0\u7ba1\u7406\u5458\u7528\u5b83\u9a8c\u8bc1\u89c6\u9891\u5185\u5bb9\u662f\u5426\u7b26\u5408\u8981\u6c42\uff0c\u6bd4\u5982\u68c0\u67e5\u662f\u5426\u6709\u8fdd\u89c4\u7247\u6bb5\u3002<\/li>\n<\/ol>\n
                                         <\/p>\n
                                        QA<\/h2>\n
                                          \n
                                        1. VideoMind\u80fd\u5904\u7406\u591a\u957f\u7684\u89c6\u9891\uff1f<\/strong>
                                          \n\u53d6\u51b3\u4e8e\u786c\u4ef6\uff0c16GB\u5185\u5b58\u53ef\u5904\u74061\u5c0f\u65f6\u89c6\u9891\uff0c32GB\u652f\u6301\u66f4\u957f\u5185\u5bb9\u3002<\/li>\n
                                        2. \u5728\u7ebf\u6f14\u793a\u548c\u672c\u5730\u90e8\u7f72\u6709\u4ec0\u4e48\u533a\u522b\uff1f<\/strong>
                                          \n\u5728\u7ebf\u7248\u65e0\u9700\u5b89\u88c5\uff0c\u4f46\u529f\u80fd\u53d7\u9650\uff1b\u672c\u5730\u7248\u652f\u6301\u81ea\u5b9a\u4e49\u8bad\u7ec3\u548c\u5927\u89c4\u6a21\u5904\u7406\u3002<\/li>\n
                                        3. \u6570\u636e\u96c6\u5982\u4f55\u4f7f\u7528\uff1f<\/strong>
                                          \n27\u4e2a\u6570\u636e\u96c6\u7528\u4e8e\u8bad\u7ec3\u548c\u8bc4\u4f30\uff0cVideoMind-SFT\u9002\u5408\u4f18\u5316\u6a21\u578b\uff0c\u5177\u4f53\u89c1https:\/\/huggingface.co\/datasets\/yeliudev\/VideoMind-Dataset\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"
                                          VideoMind \u662f\u4e00\u4e2a\u5f00\u6e90\u7684\u591a\u6a21\u6001AI\u5de5\u5177\uff0c\u4e13\u6ce8\u4e8e\u957f\u89c6\u9891\u7684\u63a8\u7406\u3001\u95ee\u7b54\u548c\u6458\u8981\u751f\u6210\u3002\u5b83\u7531\u9999\u6e2f\u7406\u5de5\u5927\u5b66\u7684\u5218\u6654\uff08Ye Liu\uff09\u548c\u65b0\u52a0\u5761\u56fd\u7acb\u5927\u5b66Show Lab\u7684\u56e2\u961f\u5171\u540c\u5f00\u53d1\u3002\u5de5\u5177\u6a21\u4eff\u4eba\u7c7b\u7406\u89e3\u89c6\u9891\u7684\u65b9\u5f0f\uff0c\u628a\u4efb\u52a1\u62c6\u6210\u89c4\u5212\u3001\u5b9a\u4f4d\u3001\u9a8c\u8bc1\u548c\u56de\u7b54\u7b49\u6b65\u9aa4\uff0c\u9010…<\/p>\n","protected":false},"author":1,"featured_media":32782,"comment_status":"open","ping_status":"closed","sticky":true,"template":"","format":"standard","meta":{"footnotes":""},"categories":[392,401,451],"tags":[230,236,197,374],"class_list":["post-29757","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-models","category-visual-recognition-model","category-video-assisted","tag-aikaiyuanxiangmu","tag-aihuiyishipinzongban","tag-aiyinshipinbianji","tag-duandaoduanyuyinzhubei"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/29757","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/comments?post=29757"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/29757\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media\/32782"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=29757"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=29757"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=29757"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}