{"id":3624,"date":"2025-07-16T21:36:47","date_gmt":"2025-07-16T13:36:47","guid":{"rendered":"https:\/\/www.15zhi.net\/blog\/?p=3624"},"modified":"2025-07-16T21:36:47","modified_gmt":"2025-07-16T13:36:47","slug":"202507-%e8%ae%ba%e6%96%87%e9%98%85%e8%af%bb-grounded-sam-assembling-open-world-models-for-diverse-visual-tasks","status":"publish","type":"post","link":"https:\/\/www.15zhi.net\/blog\/202507-%e8%ae%ba%e6%96%87%e9%98%85%e8%af%bb-grounded-sam-assembling-open-world-models-for-diverse-visual-tasks\/","title":{"rendered":"202507 \u8bba\u6587\u9605\u8bfb-Grounded SAM: Assembling Open-World Models for Diverse Visual Tasks"},"content":{"rendered":"\n

\u6765\u6e90\uff1aarXiv <\/p>\n\n\n\n

\u65f6\u95f4\uff1a2024<\/p>\n\n\n\n

\u5355\u4f4d\uff1aIDEA-Research<\/p>\n\n\n\n

\u4e00\u3001\u80cc\u666f<\/strong><\/p>\n\n\n\n

\u968f\u7740\u5f00\u653e\u4e16\u754c\u89c6\u89c9\u4efb\u52a1\u9700\u6c42\u7684\u589e\u957f\uff0c\u5982\u81ea\u52a8\u9a7e\u9a76\u3001\u673a\u5668\u4eba\u5bfc\u822a\u548c\u667a\u80fd\u5b89\u9632\u7b49\u73b0\u5b9e\u5e94\u7528\uff0c\u7cfb\u7edf\u9700\u5177\u5907\u8bc6\u522b\u4efb\u610f\u7c7b\u522b\u3001\u5904\u7406\u590d\u6742\u573a\u666f\u3001\u652f\u6301\u591a\u6a21\u6001\u8f93\u5165\u7684\u80fd\u529b\u3002\u5f53\u524d\u4e3b\u6d41\u7684\u4e09\u79cd\u65b9\u6cd5\u5206\u522b\u4e3a\uff1a<\/p>\n\n\n\n

    \n
  1. \u7edf\u4e00\u6a21\u578b\uff08Unified Models\uff09<\/strong>\uff1a\u5982 UNINEXT\u3001OFA\uff0c\u901a\u8fc7\u8bad\u7ec3\u5355\u4e2a\u6a21\u578b\u6765\u5e94\u5bf9\u591a\u79cd\u4efb\u52a1\uff0c\u4f46\u5b58\u5728\u6570\u636e\u5206\u5e03\u53d7\u9650\u3001\u5206\u5272\u80fd\u529b\u8584\u5f31\u7684\u95ee\u9898\uff1b<\/li>\n\n\n\n
  2. LLM\u63a7\u5236\u5668\u65b9\u6cd5\uff08LLM as Controller\uff09<\/strong>\uff1a\u5982 Visual ChatGPT\u3001HuggingGPT\uff0c\u901a\u8fc7\u8bed\u8a00\u5927\u6a21\u578b\u8c03\u5ea6\u89c6\u89c9\u6a21\u5757\uff0c\u4f46\u63a8\u7406\u6548\u7387\u4f4e\uff0c\u4f9d\u8d56\u6027\u5f3a\uff1b<\/li>\n\n\n\n
  3. \u57fa\u7840\u6a21\u578b\u7ec4\u88c5\uff08Ensemble Foundation Models\uff09<\/strong>\uff1a\u5c06\u591a\u4e2a\u4e13\u5bb6\u6a21\u578b\u7ec4\u5408\uff0c\u6a21\u5757\u5316\u5b8c\u6210\u590d\u6742\u4efb\u52a1\uff0c\u7075\u6d3b\u6027\u5f3a\u3002<\/li>\n<\/ol>\n\n\n\n

    \u8be5\u8bba\u6587\u91c7\u7528\u7b2c\u4e09\u79cd\u7b56\u7565\uff0c\u63d0\u51fa\u4e86\u4e00\u4e2a\u540d\u4e3a Grounded SAM<\/strong> \u7684\u7cfb\u7edf\uff0c\u65e8\u5728\u89e3\u51b3\u5f00\u653e\u8bcd\u6c47\u68c0\u6d4b\u4e0e\u5206\u5272<\/strong>\u7b49\u57fa\u7840\u89c6\u89c9\u4efb\u52a1\uff0c\u5e76\u62d3\u5c55\u81f3\u5bc6\u96c6\u6807\u6ce8\u3001\u56fe\u50cf\u7f16\u8f91\u548c\u4eba\u4f53\u5206\u6790\u7b49\u590d\u6742\u573a\u666f\u3002<\/p>\n\n\n\n

    \u4e8c\u3001\u8d21\u732e<\/strong><\/p>\n\n\n\n

      \n
    1. \u521b\u65b0\u7cfb\u7edf\u67b6\u6784<\/strong>\uff1a\u63d0\u51fa Grounded SAM<\/strong>\uff0c\u5c06 Grounding DINO\uff08\u5f00\u653e\u96c6\u68c0\u6d4b\uff09\u4e0e SAM\uff08\u53ef\u63d0\u793a\u5206\u5272\uff09\u7ed3\u5408\uff0c\u6784\u5efa\u53ef\u5904\u7406\u5f00\u653e\u8bcd\u6c47\u5206\u5272\u7684\u591a\u6a21\u578b\u7ec4\u88c5\u7ba1\u7ebf\u3002<\/li>\n\n\n\n
    2. \u6a21\u5757\u5316\u6269\u5c55\u80fd\u529b<\/strong>\uff1a\u7cfb\u7edf\u53ef\u65e0\u7f1d\u96c6\u6210 RAM\u3001BLIP\u3001Stable Diffusion\u3001OSX \u7b49\u4e13\u5bb6\u6a21\u578b\uff0c\u652f\u6301\u56fe\u50cf\u6807\u6ce8\u3001\u751f\u6210\u3001\u7f16\u8f91\u3001\u4e09\u7ef4\u4eba\u4f53\u52a8\u4f5c\u5206\u6790\u7b49\u4efb\u52a1\u3002<\/li>\n\n\n\n
    3. \u9ad8\u6548\u7ec4\u5408\u7b56\u7565<\/strong>\uff1a\u907f\u514d\u4f7f\u7528 LLM \u63a7\u5236\u5668\uff0c\u901a\u8fc7\u8bad\u7ec3\u65e0\u5173\u7684\u6a21\u578b\u7ec4\u88c5\u5b8c\u6210\u590d\u6742\u4efb\u52a1\uff0c\u5177\u5907\u8f83\u9ad8\u63a8\u7406\u6548\u7387\u4e0e\u53ef\u89e3\u91ca\u6027\u3002<\/li>\n\n\n\n
    4. \u6027\u80fd\u9886\u5148<\/strong>\uff1a\u5728 Segmentation in the Wild (SGinW) \u96f6\u6837\u672c\u8bc4\u6d4b\u4e2d\uff0cGrounded SAM \u8d85\u8d8a\u4e86 UNINEXT\u3001OpenSeeD \u7b49\u4e3b\u6d41\u65b9\u6cd5\uff0c\u9a8c\u8bc1\u5176\u5b9e\u7528\u6027\u548c\u5f3a\u6cdb\u5316\u6027\u3002<\/li>\n<\/ol>\n\n\n\n
      \"\"<\/figure>\n\n\n\n

      \u4e09\u3001\u7cfb\u7edf\u6784\u5efa\u4e0e\u573a\u666f\u5e94\u7528<\/strong><\/p>\n\n\n\n

      1. Grounded SAM \u7ec4\u6210\u6a21\u5757\uff1a<\/strong><\/p>\n\n\n\n

        <\/ol>\n\n\n\n
          \n
        • Grounding DINO<\/strong>\uff1a\u5f00\u653e\u8bcd\u6c47\u68c0\u6d4b\u5668\uff0c\u6839\u636e\u6587\u672c\u63d0\u793a\u68c0\u6d4b\u76ee\u6807\u8fb9\u754c\u6846\u3002<\/li>\n\n\n\n
        • SAM<\/strong>\uff1a\u53ef\u63d0\u793a\u7684\u5206\u5272\u6a21\u578b\uff0c\u63a5\u6536\u8fb9\u754c\u6846\u63d0\u793a\uff0c\u8f93\u51fa\u76ee\u6807\u63a9\u7801\u3002<\/li>\n\n\n\n
        • RAM<\/strong>\uff1a\u5f3a\u5927\u7684\u56fe\u50cf\u6807\u7b7e\u6a21\u578b\uff0c\u8bc6\u522b\u56fe\u50cf\u4e2d\u5e38\u89c1\u7c7b\u522b\u3002<\/li>\n\n\n\n
        • BLIP<\/strong>\uff1a\u56fe\u6587\u7406\u89e3\u4e0e\u751f\u6210\u6a21\u578b\uff0c\u751f\u6210\u56fe\u50cf\u63cf\u8ff0\u3002<\/li>\n\n\n\n
        • Stable Diffusion<\/strong>\uff1a\u56fe\u50cf\u751f\u6210\u4e0e\u4fee\u590d\u6a21\u578b\uff0c\u7528\u4e8e\u7f16\u8f91\u3001\u64e6\u9664\u7b49\u3002<\/li>\n\n\n\n
        • OSX<\/strong>\uff1a\u5168\u8eab\u7f51\u683c\u6062\u590d\u6a21\u578b\uff0c\u5b9e\u73b0\u7cbe\u786e\u4e09\u7ef4\u4eba\u4f53\u5206\u6790\u3002<\/li>\n<\/ul>\n\n\n\n

          2. \u5178\u578b\u5e94\u7528\u793a\u4f8b\uff1a<\/strong><\/p>\n\n\n\n

            \n
          • \u5f00\u653e\u96c6\u68c0\u6d4b\u4e0e\u5206\u5272\uff08Grounded SAM\uff09<\/strong>\uff1a\u5c06\u6587\u672c \u2192 \u6846 \u2192 \u63a9\u7801\u89e3\u8026\uff0c\u63d0\u5347\u5206\u5272\u51c6\u786e\u6027\u4e0e\u7075\u6d3b\u6027\uff1b<\/li>\n\n\n\n
          • \u81ea\u52a8\u56fe\u50cf\u6807\u6ce8\uff08RAM-Grounded-SAM\uff09<\/strong>\uff1a\u7528 RAM \u751f\u6210\u6807\u7b7e\u8f93\u5165 DINO+SAM \u5b9e\u73b0\u56fe\u50cf\u5168\u81ea\u52a8\u6807\u6ce8\uff1b<\/li>\n\n\n\n
          • \u56fe\u50cf\u7f16\u8f91\uff08Grounded-SAM-SD\uff09<\/strong>\uff1a\u7ed3\u5408 Stable Diffusion \u5b9e\u73b0\u5c40\u90e8\u64e6\u9664\u3001\u91cd\u7ed8\u3001\u66ff\u6362\uff1b<\/li>\n\n\n\n
          • \u4e09\u7ef4\u52a8\u4f5c\u5206\u6790\uff08Grounded-SAM-OSX\uff09<\/strong>\uff1a\u6309\u6587\u672c\u63d0\u793a\u5b9a\u4f4d\u7279\u5b9a\u4eba\u5e76\u8fdb\u884c\u7f51\u683c\u6062\u590d\uff0c\u8fdb\u884c\u5b9e\u4f8b\u5316\u52a8\u4f5c\u5206\u6790\uff1b<\/li>\n\n\n\n
          • \u591a\u6a21\u578b\u8054\u5408\u62d3\u5c55<\/strong>\uff1a\u4e0e FastSAM\u3001HQ-SAM\u3001LaMa\u3001DEVA \u7b49\u7ec4\u5408\uff0c\u52a0\u901f\u63a8\u7406\u3001\u63d0\u5347\u63a9\u7801\u8d28\u91cf\u6216\u652f\u6301\u76ee\u6807\u8ddf\u8e2a\u3002<\/li>\n<\/ul>\n\n\n\n

            \u56db\u3001\u5b9e\u9a8c<\/strong><\/p>\n\n\n\n

              \n
            • \u8bc4\u4f30\u5e73\u53f0<\/strong>\uff1aSegmentation in the Wild (SGinW)\uff0c\u5171\u5305\u542b 25 \u4e2a\u96f6\u6837\u672c\u6570\u636e\u96c6\uff1b<\/li>\n\n\n\n
            • \u8bc4\u4f30\u4efb\u52a1<\/strong>\uff1a\u5f00\u653e\u96c6\u76ee\u6807\u68c0\u6d4b + \u5206\u5272\uff1b<\/li>\n<\/ul>\n\n\n\n
              \"\"<\/figure>\n\n\n\n

              \u4e94\u3001\u603b\u7ed3<\/strong><\/p>\n\n\n\n

                \n
              • Grounded SAM \u6253\u7834\u4e86\u5355\u4e00\u6a21\u578b\u6216\u4f9d\u8d56\u8bed\u8a00\u6a21\u578b\u8c03\u5ea6\u7684\u9650\u5236\uff0c\u4ee5\u8bad\u7ec3\u65e0\u5173\u3001\u6a21\u5757\u5316\u7ec4\u88c5\u7684\u65b9\u5f0f\u5e94\u5bf9\u5f00\u653e\u4e16\u754c\u89c6\u89c9\u4efb\u52a1\uff1b<\/li>\n\n\n\n
              • \u7cfb\u7edf\u5177\u5907\u9ad8\u5ea6\u7075\u6d3b\u6027\u4e0e\u53ef\u6269\u5c55\u6027\uff0c\u9002\u7528\u4e8e\u56fe\u50cf\u6807\u6ce8\u3001\u7f16\u8f91\u3001\u4eba\u4f53\u5206\u6790\u7b49\u591a\u4e2a\u4e0b\u6e38\u4efb\u52a1\uff1b<\/li>\n\n\n\n
              • \u5728\u516c\u5f00\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8868\u73b0\u51fa\u9886\u5148\u6027\u80fd\uff0c\u5177\u5907\u6781\u5f3a\u7684\u5b9e\u7528\u4ef7\u503c\u548c\u5de5\u7a0b\u6f5c\u529b\u3002<\/li>\n<\/ul>\n\n\n\n

                <\/p>\n","protected":false},"excerpt":{"rendered":"

                \u6765\u6e90\uff1aarXiv \u65f6\u95f4\uff1a2024 \u5355\u4f4d\uff1aIDEA-Research \u4e00\u3001\u80cc\u666f \u968f\u7740\u5f00\u653e\u4e16\u754c\u89c6\u89c9\u4efb\u52a1\u9700\u6c42\u7684\u589e\u957f\uff0c […]<\/p>\n","protected":false},"author":64,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-3624","post","type-post","status-publish","format-standard","hentry","category-events"],"_links":{"self":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/3624","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\/64"}],"replies":[{"embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/comments?post=3624"}],"version-history":[{"count":3,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/3624\/revisions"}],"predecessor-version":[{"id":3662,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/posts\/3624\/revisions\/3662"}],"wp:attachment":[{"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/media?parent=3624"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/categories?post=3624"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.15zhi.net\/blog\/wp-json\/wp\/v2\/tags?post=3624"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}