{"id":1146050,"date":"2025-07-30T20:49:30","date_gmt":"2025-07-31T03:49:30","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-blog-post&p=1146050"},"modified":"2025-07-30T20:50:34","modified_gmt":"2025-07-31T03:50:34","slug":"ui-e2i-synth","status":"publish","type":"msr-blog-post","link":"https:\/\/www.microsoft.com\/en-us\/research\/articles\/ui-e2i-synth\/","title":{"rendered":"UI-E2I-Synth\uff1a\u4e3aComputer Use Agent\u6253\u9020\u66f4\u771f\u5b9e\u3001\u66f4\u5177\u6311\u6218\u6027\u7684\u5f00\u6e90Grounding\u6d4b\u8bd5\u57fa\u51c6"},"content":{"rendered":"\n
\u7f16\u8005\u6309\uff1a\u8fd1\u5e74\u6765\uff0c\u5927\u6a21\u578b\u5728\u591a\u6a21\u6001\u4ea4\u4e92\u4e2d\u7684\u5e94\u7528\u4e0d\u65ad\u6df1\u5165\uff0c\u4f7f\u5f97\u4eba\u673a\u4ea4\u4e92\u4e2d\u7684\u201c\u7406\u89e3\u80fd\u529b\u201d\u9762\u4e34\u65b0\u7684\u6311\u6218\u3002\u7279\u522b\u662f\u5728 Compute Use Agent \u7684\u53d1\u5c55\u80cc\u666f\u4e0b\uff0c\u5982\u4f55\u51c6\u786e\u7406\u89e3\u7528\u6237\u7684\u81ea\u7136\u8bed\u8a00\u6307\u4ee4\u5e76\u5c06\u5176\u6620\u5c04\u5230\u590d\u6742 GUI \u754c\u9762\u4e2d\u7684\u76f8\u5e94\u5143\u7d20\uff08\u5373 \u201cGUI Grounding\u201d \u4efb\u52a1\uff09\u6210\u4e3a\u5173\u952e\u3002\u7136\u800c\uff0c\u73b0\u6709\u7684\u516c\u5f00\u6570\u636e\u96c6\u96be\u4ee5\u5168\u9762\u8986\u76d6\u771f\u5b9e\u4f7f\u7528\u573a\u666f\u4e2d\u7684\u5404\u79cd\u96be\u70b9\u3002\u5bf9\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u63d0\u51fa\u4e86 UI-E2I-Synth\uff0c\u4e0d\u4ec5\u6784\u5efa\u4e86\u5927\u89c4\u6a21\u9ad8\u8d28\u91cf\u5408\u6210\u6570\u636e\u96c6\uff0c\u8fd8\u63a8\u51fa\u4e86\u66f4\u5177\u6311\u6218\u6027\u4e0e\u73b0\u5b9e\u6027\u7684\u6d4b\u8bd5\u57fa\u51c6 UI-I2E-Bench\uff0c\u7cfb\u7edf\u6027\u89e3\u51b3\u4e86\u5f53\u524d\u4efb\u52a1\u4e2d\u7684\u4e09\u5927\u5173\u952e\u95ee\u9898\u3002\u76f8\u5173\u8bba\u6587\u5df2\u88ab ACL 2025 \u6536\u5f55\u3002<\/p>\n\n\n\n
\u968f\u7740\u591a\u6a21\u6001\u667a\u80fd\u4f53\u5728\u590d\u6742\u4eba\u673a\u4ea4\u4e92\u4efb\u52a1\u4e2d\u7684\u5e94\u7528\u4e0d\u65ad\u88ab\u62d3\u5c55\uff0c\u5927\u6a21\u578b\u7406\u89e3\u548c\u64cd\u4f5c\u56fe\u5f62\u7528\u6237\u754c\u9762\uff08GUI\uff09\u7684\u80fd\u529b\u6b63\u53d8\u5f97\u65e5\u76ca\u5173\u952e\u3002GUI Grounding\uff0c\u5373\u6839\u636e\u7528\u6237\u7684\u81ea\u7136\u8bed\u8a00\u6307\u4ee4\uff0c\u51c6\u786e\u5b9a\u4f4d\u5c4f\u5e55\u622a\u56fe\u4e2d\u5bf9\u5e94\u7684\u754c\u9762\u5143\u7d20\uff0c\u5df2\u6210\u4e3a\u5f53\u524d Compute Use Agent \u5f00\u53d1\u4e2d\u7684\u6838\u5fc3\u6311\u6218\u4e4b\u4e00\u3002<\/p>\n\n\n\n
\u7136\u800c\uff0c\u73b0\u6709\u7684\u516c\u5f00\u8bad\u7ec3\u6570\u636e\u96c6\u89c4\u6a21\u6709\u9650\uff0c\u4e14\u9ad8\u8d28\u91cf\u6307\u4ee4\u7684\u4eba\u5de5\u6807\u6ce8\u6210\u672c\u9ad8\u6602\uff0c\u5236\u7ea6\u4e86\u8be5\u4efb\u52a1\u7684\u8fdb\u4e00\u6b65\u53d1\u5c55\u3002\u5f53\u524d\u7814\u7a76\u9762\u4e34\u7740\u4e09\u4e2a\u5c1a\u672a\u88ab\u5145\u5206\u63a2\u7d22\u7684\u5173\u952e\u95ee\u9898\uff1a<\/p>\n\n\n\n
1. \u5143\u7d20\u4e0e\u5c4f\u5e55\u7684\u6bd4\u4f8b\u5931\u771f<\/strong>\uff1a\u73b0\u6709\u57fa\u51c6\u6d4b\u8bd5\uff08\u5982ScreenSpot\uff09\u7684\u754c\u9762\u5143\u7d20\u76f8\u5bf9\u5c4f\u5e55\u660e\u663e\u504f\u5927\uff0c\u8fdc\u79bb\u771f\u5b9e\u7528\u6237\u573a\u666f\u4e2d\u7684\u9ad8\u5206\u8fa8\u7387\u3001\u5c0f\u5143\u7d20\u754c\u9762\u3002\u8fd9\u53ef\u80fd\u5bfc\u81f4\u6a21\u578b\u7684\u6027\u80fd\u88ab\u9ad8\u4f30\u3002<\/p>\n\n\n\n 2. \u5143\u7d20\u7c7b\u578b\u5206\u5e03\u5931\u8861<\/strong>\uff1a\u4e0d\u540c\u7c7b\u578b\u7684 GUI \u5143\u7d20\uff08\u5982\u6587\u672c\u6309\u94ae\u3001\u590d\u9009\u6846\u7b49\uff09\u5177\u6709\u4e0d\u540c\u7684\u89c6\u89c9\u98ce\u683c\u548c\u4ea4\u4e92\u65b9\u5f0f\u3002\u4f46\u73b0\u6709\u57fa\u51c6\u4e2d\u6587\u672c\u548c\u56fe\u6807\u7c7b\u578b\u5360\u6bd4\u8fc7\u9ad8\uff0c\u7f3a\u4e4f\u5bf9\u5176\u4ed6\u7c7b\u578b\u7684\u8986\u76d6\u3002<\/p>\n\n\n\n 3. \u9690\u5f0f\u6307\u4ee4\u7406\u89e3\u80fd\u529b\u5f31<\/strong>\uff1a\u73b0\u5b9e\u7528\u6237\u5728\u53d1\u51fa\u6307\u4ee4\u65f6\u5f80\u5f80\u4f9d\u8d56\u4e8e\u5bf9\u5143\u7d20\u529f\u80fd\u6216\u4f4d\u7f6e\u7684\u76f4\u89c2\u7406\u89e3\uff0c\u800c\u975e\u76f4\u63a5\u5f15\u7528\u5c4f\u5e55\u4e0a\u7684\u53ef\u89c1\u6587\u672c\uff0c\u8fd9\u7c7b\u201c\u9690\u5f0f\u6307\u4ee4\u201d\u5bf9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\u7684\u7406\u89e3\u548c\u63a8\u7406\u80fd\u529b\u63d0\u51fa\u4e86\u66f4\u9ad8\u7684\u8981\u6c42\u3002<\/p>\n\n\n\n \u9488\u5bf9\u8fd9\u4e9b\u95ee\u9898\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u7aef\u5230\u7aef\u7684\u5927\u89c4\u6a21\u6570\u636e\u5408\u6210\u65b9\u6cd5 UI-E2I-Synth\uff0c\u5e76\u4f7f\u7528\u534a\u81ea\u52a8\u65b9\u5f0f\u6784\u5efa\u4e86\u4e00\u4e2a\u5168\u65b0\u7684\u6d4b\u8bd5\u57fa\u51c6 UI-I2E-Bench\u3002\u76f8\u5173\u8bba\u6587\u5df2\u88ab ACL 2025 \u63a5\u6536\u3002<\/p>\n\n\n\n UI-E2I-Synth: Advancing GUI Grounding with Large-Scale Instruction Synthesis<\/strong><\/p>\n\n\n\n \u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/2504.11257 (opens in new tab)<\/span><\/a><\/p>\n\n\n\n GitHub \u94fe\u63a5\uff1ahttps:\/\/github.com\/microsoft\/FIVE-UI-Evol\/tree\/main (opens in new tab)<\/span><\/a><\/p>\n\n\n\n