{"id":36589,"date":"2025-01-14T14:00:29","date_gmt":"2025-01-14T06:00:29","guid":{"rendered":"https:\/\/17aitech.com\/?p=36589"},"modified":"2025-01-14T14:00:29","modified_gmt":"2025-01-14T06:00:29","slug":"%e4%b8%8d%e8%ae%a9%e8%a7%86%e8%a7%89%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b%e3%80%8c%e7%9b%b2%e7%8c%9c%e3%80%8d%ef%bc%8c%e6%80%a7%e8%83%bd%e7%ab%9f%e7%9b%b4%e6%8e%a5%e6%8f%90%e5%8d%87%e4%b8%80%e5%80%8d","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=36589","title":{"rendered":"\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2024-11-09\" target=\"_blank\">\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f<\/a><\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-1efd6e82a73a4a96cc4bd7d1d971a947.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-1efd6e82a73a4a96cc4bd7d1d971a947.png\"><\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"166\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<p>AIxiv\u4e13\u680f\u662f\u673a\u5668\u4e4b\u5fc3\u53d1\u5e03\u5b66\u672f\u3001\u6280\u672f\u5185\u5bb9\u7684\u680f\u76ee\u3002\u8fc7\u53bb\u6570\u5e74\uff0c\u673a\u5668\u4e4b\u5fc3AIxiv\u4e13\u680f\u63a5\u6536\u62a5\u9053\u4e862000\u591a\u7bc7\u5185\u5bb9\uff0c\u8986\u76d6\u5168\u7403\u5404\u5927\u9ad8\u6821\u4e0e\u4f01\u4e1a\u7684\u9876\u7ea7\u5b9e\u9a8c\u5ba4\uff0c\u6709\u6548\u4fc3\u8fdb\u4e86\u5b66\u672f\u4ea4\u6d41\u4e0e\u4f20\u64ad\u3002\u5982\u679c\u60a8\u6709\u4f18\u79c0\u7684\u5de5\u4f5c\u60f3\u8981\u5206\u4eab\uff0c\u6b22\u8fce\u6295\u7a3f\u6216\u8005\u8054\u7cfb\u62a5\u9053\u3002\u6295\u7a3f\u90ae\u7bb1\uff1aliyazhou@jiqizhixin.com\uff1bzhaoyunfeng@jiqizhixin.com<\/p>\n<\/section>\n<\/blockquote>\n<p><strong>\u56e2\u961f\u7684\u5171\u540c\u4e00\u4f5c\u6797\u4e4b\u79cb\uff08Zhiqiu Lin\uff09\u662f\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\u7684\u535a\u58eb\u7814\u7a76\u751f\uff0c\u7531Deva Ramanan\u6559\u6388\u6307\u5bfc\uff0c\u4e13\u6ce8\u4e8e\u89c6\u89c9-\u8bed\u8a00\u5927\u6a21\u578b\u7684\u81ea\u52a8\u8bc4\u4f30\u4e0e\u4f18\u5316\u3002\u56e2\u961f\u7684\u9a6c\u7d2b\u665b\uff08Zixian Ma\uff09\u662f\u534e\u76db\u987f\u5927\u5b66\u7684\u535a\u58eb\u7814\u7a76\u751f\uff0c\u7531Ranjay Krishna\u6559\u6388\u6307\u5bfc\uff0c\u4e13\u6ce8\u4e8e\u89c6\u89c9-\u8bed\u8a00\u5927\u6a21\u578b\u7684\u8bc4\u6d4b\u4ee5\u53ca\u4e0e\u4eba\u7684\u4ea4\u4e92\u3002<\/strong><\/p>\n<section>\u5728\u8bf8\u5982 MME\u3001MMBench\u3001MMMU \u548c ScienceQA \u7b49\u590d\u6742\u7684\u89c6\u89c9\u95ee\u7b54\uff08VQA\uff09\u57fa\u51c6\u4e0a\uff0cGPT-4o\u3001Qwen2-VL \u7b49\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u5df2\u7ecf\u53d6\u5f97\u4e86\u663e\u8457\u8fdb\u5c55\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u6a21\u578b\u771f\u7684\u7406\u89e3\u81ea\u7136\u56fe\u50cf\u5417\uff1f<\/section>\n<section><\/section>\n<section>\u8fd1\u65e5\uff0c\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\u4e0e\u534e\u76db\u987f\u5927\u5b66\u7684\u7814\u7a76\u56e2\u961f\u63a8\u51fa\u4e86 NaturalBench\uff0c\u8fd9\u662f\u4e00\u9879\u53d1\u8868\u4e8e NeurIPS&#8217;24 \u7684\u4ee5\u89c6\u89c9\u4e3a\u6838\u5fc3\u7684 VQA \u57fa\u51c6\u3002\u5b83\u901a\u8fc7\u81ea\u7136\u56fe\u50cf\u4e0a\u7684\u7b80\u5355\u95ee\u9898\u2014\u2014\u5373\u81ea\u7136\u5bf9\u6297\u6837\u672c\uff08Natural Adversarial Samples\uff09\u2014\u2014\u5bf9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u53d1\u8d77\u4e25\u5cfb\u6311\u6218\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-82da2ee8e52643ae335746e6ac74a68a.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-82da2ee8e52643ae335746e6ac74a68a.png\"><\/a><\/section>\n<ul>\n<li>\n<section>\ud83d\udcc4\u8bba\u6587\u94fe\u63a5: https:\/\/arxiv.org\/abs\/2410.14669<\/section>\n<\/li>\n<li>\n<section>\ud83c\udf0d\u7f51\u7ad9: https:\/\/huggingface.co\/datasets\/BaiqiL\/NaturalBench<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLMs\uff09\uff0c\u5982 GPT-4o\u3001BLIP-3\u3001LLaVA-OneVision\u3001Llama3.2-Vision\u3001Molmo \u548c Qwen2-VL\uff0c\u5728 MMMU \u548c MME \u7b49\u590d\u6742\u89c6\u89c9\u95ee\u7b54\uff08VQA\uff09\u57fa\u51c6\u4e0a\u8868\u73b0\u51fa\u8272\u3002\u8fd9\u4e9b\u57fa\u51c6\u6db5\u76d6\u4e86\u5927\u5b66\u8bfe\u7a0b\u3001\u5e38\u8bc6\u63a8\u7406\u3001\u56fe\u8868\u7406\u89e3\uff0c\u4ee5\u53ca\u6570\u5b66\u3001\u7f16\u7a0b\u3001\u7269\u7406\u548c\u65f6\u5e8f\u9884\u6d4b\u7b49\u4efb\u52a1\u3002<\/section>\n<section><\/section>\n<section>\u7136\u800c\uff0c\u7814\u7a76\u56e2\u961f\u53d1\u73b0\uff0c\u8fd9\u4e9b\u6a21\u578b\u5728\u5e94\u5bf9\u81ea\u7136\u56fe\u50cf\u4e2d\u7684\u7b80\u5355\u95ee\u9898\u65f6\u4ecd\u5b58\u5728\u4e0d\u8db3\u3002\u4e0b\u56fe\u5c55\u793a\u4e86\u4eba\u7c7b\u80fd\u8f7b\u677e\u89e3\u7b54\u7684 VQA \u793a\u4f8b\uff0c\u4f46\u5373\u4f7f\u662f\u6700\u5148\u8fdb\u7684\u6a21\u578b\u4e5f\u5e38\u4f1a\u51fa\u9519\uff0c\u7814\u7a76\u56e2\u961f\u5c06\u8fd9\u4e9b\u95ee\u9898\u79f0\u4e3a\u300c\u81ea\u7136\u5bf9\u6297\u6837\u672c\u300d\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-c9be967f0054eda088ed1f32f55f57eb.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-c9be967f0054eda088ed1f32f55f57eb.png\"><\/a><\/p>\n<section>\u6a21\u578b\u4e4b\u6240\u4ee5\u80fd\u5728\u4ee5\u5f80\u6d41\u884c\u7684 VQA \u57fa\u51c6\u4e0a\u8868\u73b0\u51fa\u8272\uff0c\u4e3b\u8981\u4f9d\u8d56\u4e8e\u5176\u8fc7\u5ea6\u7684\u8bed\u8a00\u504f\u89c1\uff08Language Bias\uff09\u3002\u4e3a\u8bc1\u660e\u8fd9\u4e00\u70b9\uff0c\u7814\u7a76\u56e2\u961f\u5c55\u793a\u4e86\u516d\u5927\u57fa\u51c6\u4e2d\u7684\u4e00\u4e9b\u95ee\u9898\uff0c\u5373\u4f7f\u4e0d\u67e5\u770b\u56fe\u50cf\u4e5f\u80fd\u4f5c\u7b54\u3002<\/section>\n<section><\/section>\n<section>\u4f8b\u5982\uff0c\u6a21\u578b\u53ef\u4ee5\u901a\u8fc7\u8bed\u8a00\u5148\u9a8c\uff08Language Prior\uff09\u56de\u7b54\u8bf8\u5982\u300c\u9a6c\u8428\u8bf8\u585e\u5dde\u7684\u9996\u5e9c\u662f\u4ec0\u4e48\uff1f\u300d\uff08\u300c\u6ce2\u58eb\u987f\u300d\uff09\u548c\u300c\u56fe\u7247\u4e2d\u6709\u9ed1\u8272\u957f\u9888\u9e7f\u5417\uff1f\u300d\uff08\u300c\u5426\u300d\uff09\u8fd9\u6837\u7684\u95ee\u9898\uff0c\u800c\u65e0\u9700\u4f9d\u8d56\u56fe\u50cf\u4fe1\u606f\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0cNaturalBench \u8bbe\u8ba1\u4e86\u914d\u5bf9\u4efb\u52a1\uff0c\u5c06\u4e24\u5e45\u56fe\u7247\u4e0e\u4e24\u4e2a\u76f8\u53cd\u7b54\u6848\u7684\u95ee\u9898\u5339\u914d\uff0c\u4ee5\u907f\u514d\u6a21\u578b\u53ef\u4ee5\u51ed\u300c\u76f2\u731c\u300d\u4fa5\u5e78\u7b54\u5bf9\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-9367c9c461a7ee8b5e180f8a2524feaf.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-9367c9c461a7ee8b5e180f8a2524feaf.png\"><\/a><\/p>\n<section><strong>NaturalBench \u6570\u636e\u96c6\u6536\u96c6\u8fc7\u7a0b<\/strong><\/section>\n<section><\/section>\n<section>NaturalBench \u901a\u8fc7\u4e00\u4e2a\u7b80\u5355\u7684\u6d41\u7a0b\u4ece Flickr30K \u7b49\u56fe\u6587\u6570\u636e\u96c6\u4e2d\u6784\u5efa\uff0c\u5177\u4f53\u6b65\u9aa4\u5982\u4e0b\uff1a<\/section>\n<section><\/section>\n<section>1. \u627e\u51fa CLIP \u65e0\u6cd5\u6b63\u786e\u5339\u914d\u7684\u56fe\u6587\u5bf9\u3002<\/section>\n<section>2. \u4f7f\u7528 ChatGPT \u4e3a\u6bcf\u4e2a\u56fe\u6587\u5bf9\u751f\u6210\u76f8\u53cd\u7b54\u6848\u7684\u95ee\u9898\u3002<\/section>\n<section><\/section>\n<section>\u901a\u8fc7\u907f\u514d\u5bf9\u56fe\u50cf\u6216\u95ee\u9898\u8fdb\u884c\u975e\u81ea\u7136\u5e72\u6270\uff0cNaturalBench \u751f\u6210\u4e86\u81ea\u7136\u7684\u5bf9\u6297\u6837\u672c\u3002\u8fd9\u4e9b\u6837\u672c\u57fa\u4e8e\u81ea\u7136\u56fe\u50cf\u63d0\u51fa\u7684\u95ee\u9898\uff0c\u4eba\u7c7b\u53ef\u4ee5\u8f7b\u677e\u7406\u89e3\u5e76\u56de\u7b54\uff0c\u4f46\u5bf9\u6a21\u578b\u6765\u8bf4\u5374\u6781\u5177\u6311\u6218\u6027\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-915bbfee3ad9f1bbe8297c2af3c859af.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-915bbfee3ad9f1bbe8297c2af3c859af.png\"><\/a><\/p>\n<section>\u901a\u8fc7\u8be5\u6d41\u7a0b\u53ca\u4eba\u5de5\u8d28\u68c0\uff0c\u7814\u7a76\u56e2\u961f\u6700\u7ec8\u6536\u96c6\u4e86 10,000 \u4e2a\u9ad8\u8d28\u91cf VQA \u6837\u672c\uff0c\u7528\u4e8e\u53ef\u9760\u8bc4\u4f30\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u3002<\/section>\n<section><\/section>\n<section><strong>NaturalBench \u5b9e\u9a8c\u7ed3\u679c<\/strong><\/section>\n<section><\/section>\n<section>\u57fa\u4e8e NaturalBench \u4e2d\u300c\u4e24\u5f20\u56fe\u7247 + \u4e24\u4e2a\u95ee\u9898\u300d\u914d\u5bf9\u7684\u683c\u5f0f\uff0c\u7814\u7a76\u56e2\u961f\u63d0\u51fa\u4e86\u66f4\u53ef\u9760\u7684\u89c6\u89c9\u4e3a\u4e2d\u5fc3\u8bc4\u4f30\u6307\u6807 \u2014\u2014<strong>Group Accuracy (G-Acc)<\/strong>\u3002\u53ea\u6709\u5f53\u6a21\u578b\u6b63\u786e\u56de\u7b54\u4e00\u4e2a\u6837\u672c\u4e2d\u7684\u6240\u6709\u56db\u4e2a\uff08\u56fe\u7247\u3001\u95ee\u9898\uff09\u7ec4\u5408\u65f6\uff0c\u624d\u80fd\u5f97\u4e00\u5206\u3002<\/section>\n<section><\/section>\n<section>\u7814\u7a76\u4eba\u5458\u53d1\u73b0\u4e86\u4ee5\u4e0b\u91cd\u8981\u7684\u5b9e\u9a8c\u7ed3\u8bba\uff1a<\/section>\n<section><\/section>\n<section>1\u3001<strong>\u5148\u524d\u7684 VQA \u57fa\u51c6\u5b58\u5728\u76f2\u731c\u95ee\u9898<\/strong>\uff1a\u6d41\u884c\u7684 VQA \u57fa\u51c6\uff0c\uff08\u5982 MME \u548c ScienceQA\uff09\u53ef\u4ee5\u901a\u8fc7\u5fae\u8c03\u4e0d\u5177\u89c6\u89c9\u80fd\u529b\u7684 GPT-3.5 \u6765\u89e3\u51b3\uff0c\u5176\u8868\u73b0\u751a\u81f3\u8d85\u8fc7\u4e86\u7ecf\u8fc7\u89c6\u89c9\u5fae\u8c03\u7684 LLaVA-1.5\u3002\u7136\u800c\uff0c\u5728 NaturalBench \u4e0a\uff0c\u8fd9\u79cd\u300c\u76f2\u731c\u300d\u6a21\u578b\u53ea\u80fd\u5f97\u5230\u96f6\u5206\uff01<\/section>\n<section><\/section>\n<section>\u8fd9\u8868\u660e\uff0c\u5f53\u524d\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u53ef\u80fd\u5728\u89c6\u89c9\u95ee\u7b54\u4efb\u52a1\u4e0a\u4f9d\u8d56\u5f3a\u5927\u7684\u8bed\u8a00\u76f2\u731c\u80fd\u529b<strong>\u300c\u6d51\u6c34\u6478\u9c7c\u300d<\/strong>\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-0a7095293448d09807cb869eea1083b5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-0a7095293448d09807cb869eea1083b5.png\"><\/a><\/p>\n<section>2\u3001<strong>\u5f53\u524d\u5f00\u6e90\u6a21\u578b\u7684\u8868\u73b0\u6709\u9650<\/strong>\uff1a\u7814\u7a76\u56e2\u961f\u5bf9 53 \u4e2a\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u4e86\u6d4b\u8bd5\uff0c\u7ed3\u679c\u663e\u793a\uff0c\u5927\u591a\u6570\u5f00\u6e90\u6a21\u578b\uff08\u5982 BLIP-3\u3001Cambrian-1\u3001LLaVA-OneVision\u3001Llama3.2-Vision\u3001Molmo \u548c Qwen2-VL\uff09\u7684\u5f97\u5206\u4ec5\u6bd4\u968f\u673a\u6c34\u5e73\u9ad8 10%-20%\u3002\u5373\u4f7f\u662f\u7ecf\u8fc7\u89c6\u89c9\u5fae\u8c03\u7684 GPT-4o\uff0c\u5176\u8868\u73b0\u4ecd\u6bd4\u4eba\u7c7b\u4f4e 50%\u3002<\/section>\n<section><\/section>\n<section>\u8fd9\u8868\u660e\uff0c\u76ee\u524d\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u5728\u81ea\u7136\u56fe\u50cf\u7684\u89c6\u89c9\u7406\u89e3\u80fd\u529b\u4e0a\u8fd8\u6709\u5de8\u5927\u63d0\u5347\u7a7a\u95f4\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-db477c966a24ae8899875b035e004a05.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-db477c966a24ae8899875b035e004a05.png\"><\/a><\/p>\n<section><strong>NaturalBench \u5b9e\u9a8c\u8ba8\u8bba<\/strong><\/section>\n<section><\/section>\n<section>\u7814\u7a76\u4eba\u5458\u5bf9\u63d0\u5347\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u7684\u65b9\u5411\u8fdb\u884c\u4e86\u63a2\u8ba8\uff0c\u5e76\u53d1\u73b0\u4e24\u4e2a\u91cd\u70b9\uff1a<\/section>\n<section><\/section>\n<section>1\u3001\u8ba9\u5927\u6a21\u578b\u4e0d\u518d\u300c\u76f2\u9009\u300d\uff1a\u5927\u591a\u6570\u6a21\u578b\u5931\u8d25\u7684\u5173\u952e\u5728\u4e8e\uff0c\u65e0\u8bba\u56fe\u50cf\u5185\u5bb9\u5982\u4f55\uff0c\u5b83\u4eec\u603b\u662f\u76f2\u76ee\u9009\u62e9\u76f8\u540c\u7684\u9009\u9879\u3002\u5728 GPT-4o \u51fa\u9519\u7684\u95ee\u9898\u4e2d\uff0c\u8d85\u8fc7 80% \u662f\u56e0\u4e3a\u6a21\u578b\u5728\u4efb\u4f55\u56fe\u50cf\u4e0b\u90fd\u9009\u62e9\u4e86\u540c\u4e00\u7b54\u6848\uff08\u5982\u300c\u662f\u300d\uff09\u3002\u7814\u7a76\u4eba\u5458\u53d1\u73b0\uff0c\u7ea0\u6b63\u8fd9\u4e00\u503e\u5411\u540e\uff0c\u6a21\u578b\u6027\u80fd\u53ef\u63d0\u5347\u4e24\u5230\u4e09\u500d\uff01<\/section>\n<section><\/section>\n<section>\u7814\u7a76\u56e2\u961f\u91c7\u7528\u4e86\u4e00\u79cd\u57fa\u4e8e\u8bc4\u5206\u7684\u8bc4\u4f30\u65b9\u5f0f\uff08VQAScore\uff09\uff0c\u901a\u8fc7\u8c03\u6574\u540c\u4e00\u95ee\u9898\u4e0b\u4e24\u4e2a\u4e0d\u540c\u9009\u9879\u7684\u5f97\u5206\u5dee\uff0c\u786e\u4fdd\u6a21\u578b\u5728\u56de\u7b54\u300c\u662f\u300d\u548c\u300c\u5426\u300d\u65f6\u7684\u6bd4\u4f8b\u4fdd\u6301\u4e00\u81f4\u3002\u4ec5\u51ed\u8fd9\u4e00\u7b80\u5355\u8c03\u6574\uff0c\u50cf GPT-4o \u8fd9\u6837\u7684\u5f3a\u5927\u6a21\u578b\u7684\u8868\u73b0\u51e0\u4e4e\u7ffb\u4e86\u4e24\u500d\uff01<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-eec7189279e6f69e974bb36822ad2aa8.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-eec7189279e6f69e974bb36822ad2aa8.png\"><\/a><\/p>\n<section>\u8fd9\u610f\u5473\u7740 <strong>NaturalBench<\/strong> \u53ef\u4ee5\u4f5c\u4e3a\u4e00\u4e2a\u7ea0\u6b63\u6a21\u578b\u504f\u89c1\u7684\u57fa\u51c6\uff0c\u5e2e\u52a9\u51cf\u5c11\u6a21\u578b\u7684\u76f2\u731c\u884c\u4e3a\u548c\u5e7b\u89c9\u95ee\u9898\u3002<\/section>\n<section><\/section>\n<section>2\u3001\u5927\u6a21\u578b\u4ecd\u9700\u63d0\u5347\u7ec4\u5408\u6027\u601d\u7ef4\uff08compositional reasoning\uff09\u80fd\u529b\uff1aNaturalBench \u7684\u5927\u90e8\u5206\u95ee\u9898\u8981\u6c42\u6a21\u578b\u540c\u65f6\u5177\u5907\u591a\u79cd\u89c6\u89c9\u6280\u80fd\uff08compositional reasoning\uff09\uff0c\u5305\u62ec\u5bf9\u8c61\uff08object\uff09\u3001\u5c5e\u6027\uff08attribute\uff09\u3001\u5173\u7cfb\uff08relation\uff09\u548c\u903b\u8f91\uff08logic\uff09\u7b49\u7ef4\u5ea6\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u6b64\uff0c\u7814\u7a76\u56e2\u961f\u4e3a NaturalBench \u4e2d\u7684\u6bcf\u4e2a\u95ee\u9898\u6807\u6ce8\u4e86 1 \u5230 8 \u4e2a\u6280\u80fd\u6807\u7b7e\uff0c\u5bf9\u7ec4\u5408\u63a8\u7406\u80fd\u529b\u8fdb\u884c\u4e86\u7ec6\u81f4\u8bc4\u4f30\u3002\u7ed3\u679c\u8868\u660e\uff0c\u5373\u4fbf\u662f GPT-4o\uff0c\u5728\u7a7a\u95f4\u5173\u7cfb\u3001\u903b\u8f91\u63a8\u7406\u7b49\u65b9\u9762\u4ecd\u6709\u663e\u8457\u7684\u63d0\u5347\u7a7a\u95f4\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-fa9e1630caaf32666ecdbaa27afc48f7.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-fa9e1630caaf32666ecdbaa27afc48f7.png\"><\/a><\/p>\n<section><strong>NaturalBench \u5bf9\u52a8\u6001\u8bc4\u6d4b\uff08Dynamic Evaluation\uff09\u7684\u610f\u4e49<\/strong><\/section>\n<section><\/section>\n<section>\u4f20\u7edf\u673a\u5668\u5b66\u4e60\u6a21\u578b\u5728\u9759\u6001\u57fa\u51c6\u4e0a\u8ffd\u5e73\u4eba\u7c7b\u8868\u73b0\u901a\u5e38\u9700\u8981\u591a\u5e74\uff0c\u4f8b\u5982 MNIST \u7528\u4e86 15 \u5e74\uff0cImageNet \u7528\u4e86 7 \u5e74\u3002\u7136\u800c\uff0c\u7531\u4e8e\u5f53\u524d\u5927\u6a21\u578b\u5927\u91cf\u4f7f\u7528\u4e92\u8054\u7f51\u6570\u636e\uff0c\u6570\u636e\u96c6\u6cc4\u9732\u4f7f\u65b0\u57fa\u51c6\u5728\u6570\u6708\u5185\u5c31\u53ef\u80fd\u8fc7\u65f6\u3002\u56e0\u6b64\uff0c\u8fd1\u671f\u7814\u7a76\u63d0\u5021<strong>\u52a8\u6001\u8bc4\u4f30\uff08Dynamic Evaluation\uff09<\/strong>\uff0c\u5e38\u89c1\u7684\u65b9\u6cd5\u662f\u901a\u8fc7\u4eba\u673a\u534f\u4f5c\uff08human-and-model-in-the-loop\uff09\u6536\u96c6\u5bf9\u6297\u6837\u672c\u3002<\/section>\n<section><\/section>\n<section>\u4f8b\u5982\uff0cAdversarial NLI \u548c Dynabench \u8981\u6c42\u6807\u6ce8\u8005\u4e0d\u65ad\u8bbe\u8ba1\u96be\u9898\u4f7f\u6a21\u578b\u51fa\u9519\uff0c\u800c Adversarial VQA \u8ba9\u6807\u6ce8\u8005\u4e3a\u56fe\u7247\u53cd\u590d\u751f\u6210\u95ee\u9898\uff0c\u76f4\u5230\u6a21\u578b\u5931\u8d25\u3002<\/section>\n<section><\/section>\n<section>\u76f8\u6bd4\u4e4b\u4e0b\uff0cNaturalBench\u00a0<strong>\u4e0d\u9488\u5bf9\u7279\u5b9a VQA \u6a21\u578b<\/strong>\uff0c\u4e14\u53ea\u9700\u6807\u6ce8\u8005<strong>\u4e00\u6b21\u6027\u9a8c\u8bc1<\/strong>\uff0c\u5927\u5927\u63d0\u9ad8\u4e86\u52a8\u6001\u57fa\u51c6\u6784\u5efa\u7684\u6548\u7387\u3002\u7814\u7a76\u56e2\u961f\u8fd8\u4f7f\u7528\u5148\u8fdb\u7684 LongCLIP \u7b49\u6a21\u578b\uff0c\u4ece DOCCI \u548c XM3600 \u6570\u636e\u96c6\u4e2d\u6536\u96c6\u4e86\u66f4\u590d\u6742\u4e14\u591a\u8bed\u79cd\u7684 VQA \u6837\u672c\uff0c\u8bc1\u660e\u4e86 NaturalBench \u5728\u672a\u6765\u5927\u6a21\u578b\u7684\u52a8\u6001\u8bc4\u4f30\u4e2d\u5177\u6709\u5e7f\u6cdb\u7684\u9002\u7528\u6027\u3002<\/section>\n<section><\/section>\n<section>\u6700\u540e\uff0cNaturalBench \u6570\u636e\u96c6\u5df2\u7ecf\u5f00\u6e90\uff1ahttps:\/\/huggingface.co\/datasets\/BaiqiL\/NaturalBench\/\u00a0<\/section>\n<section><\/section>\n<section>\u671f\u5f85\u672a\u6765\u66f4\u52a0\u5f3a\u5927\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u95ee\u4e16\uff01<\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2024-11-09\" target=\"_blank\">\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[27],"tags":[68,55],"class_list":["post-36589","post","type-post","status-publish","format-standard","hentry","category-news","tag-68","tag-55"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f - \u4e00\u8d77AI\u6280\u672f<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/17aitech.com\/?p=36589\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=36589\",\"url\":\"https:\/\/17aitech.com\/?p=36589\",\"name\":\"\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=36589#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=36589#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-1efd6e82a73a4a96cc4bd7d1d971a947.png\",\"datePublished\":\"2025-01-14T06:00:29+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=36589#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=36589\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=36589#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-1efd6e82a73a4a96cc4bd7d1d971a947.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-1efd6e82a73a4a96cc4bd7d1d971a947.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=36589#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/17aitech.com\/#website\",\"url\":\"https:\/\/17aitech.com\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/17aitech.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"url\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"contentUrl\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\/\/17aitech.com\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f - \u4e00\u8d77AI\u6280\u672f","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/17aitech.com\/?p=36589","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=36589","url":"https:\/\/17aitech.com\/?p=36589","name":"\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=36589#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=36589#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-1efd6e82a73a4a96cc4bd7d1d971a947.png","datePublished":"2025-01-14T06:00:29+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=36589#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=36589"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=36589#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-1efd6e82a73a4a96cc4bd7d1d971a947.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-1efd6e82a73a4a96cc4bd7d1d971a947.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=36589#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u4e0d\u8ba9\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u300c\u76f2\u731c\u300d\uff0c\u6027\u80fd\u7adf\u76f4\u63a5\u63d0\u5347\u4e00\u500d\uff1f"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/36589","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=36589"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/36589\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=36589"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=36589"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=36589"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}