{"id":26486,"date":"2025-03-19T17:21:04","date_gmt":"2025-03-19T09:21:04","guid":{"rendered":"http:\/\/192.168.10.115\/?p=26486"},"modified":"2025-03-19T17:21:04","modified_gmt":"2025-03-19T09:21:04","slug":"2025-03-19-%e8%87%aa%e5%8a%a8%e9%a9%be%e9%a9%b6%e8%ae%ba%e6%96%87%e9%80%9f%e9%80%92-dynrsl-vlm%e3%80%81centaur%e3%80%81drivegen%e3%80%81sparsealign%e7%ad%89%e6%9c%80%e6%96%b0%e5%b7%a5%e4%bd%9c","status":"publish","type":"post","link":"http:\/\/222.128.65.89:18086\/index.php\/2025\/03\/19\/26486\/","title":{"rendered":"2025-03-19 \u81ea\u52a8\u9a7e\u9a76\u8bba\u6587\u901f\u9012 | DynRsl-VLM\u3001Centaur\u3001DriveGEN\u3001SparseAlign\u7b49\u6700\u65b0\u5de5\u4f5c\uff082025.3.19\uff09"},"content":{"rendered":"\n<p>\u539f\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/zhuanlan.zhihu.com\/p\/31173300641\">https:\/\/zhuanlan.zhihu.com\/p\/31173300641<\/a><\/p>\n\n\n\n<p>\u4eca\u65e5\u8bba\u6587\u76d8\u70b9\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7aef\u5230\u7aef\u65b0\u7b97\u6cd5<a href=\"https:\/\/zhida.zhihu.com\/search?content_id=255257139&amp;content_type=Article&amp;match_order=1&amp;q=Centaur&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">Centaur<\/a><\/li>\n\n\n\n<li>\u81ea\u52a8\u9a7e\u9a76VLM DynRsl-VLM\uff1b<\/li>\n\n\n\n<li>\u57fa\u4e8e\u6269\u6563\u6a21\u578b\u76843D\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5DriveGEN\uff1b<\/li>\n\n\n\n<li>\u7a00\u758f\u611f\u77e5<a href=\"https:\/\/zhida.zhihu.com\/search?content_id=255257139&amp;content_type=Article&amp;match_order=1&amp;q=SparseAlign&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">SparseAlign<\/a><\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_31173300641_0\">Centaur<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aCentaur: Robust End-to-End Autonomous Driving with <a href=\"https:\/\/zhida.zhihu.com\/search?content_id=255257139&amp;content_type=Article&amp;match_order=1&amp;q=Test-Time+Training&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">Test-Time Training<\/a><\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.11650\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.11650<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-59632a62f52efa0c9256bb9eb263dfbc_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u6d4b\u8bd5\u65f6\u8bad\u7ec3\uff08Test-Time Training, TTT\uff09\u5728\u7aef\u5230\u7aef\u81ea\u52a8\u9a7e\u9a76\u4e2d\u7684\u9996\u6b21\u5e94\u7528<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u51fa\u4e00\u79cd\u5728\u7ebf\u4f18\u5316\u6846\u67b6\uff0c\u901a\u8fc7\u5728\u90e8\u7f72\u9636\u6bb5\u5bf9\u6a21\u578b\u53c2\u6570\u8fdb\u884c\u5355\u6b65\u68af\u5ea6\u66f4\u65b0\uff0c<strong>\u6700\u5c0f\u5316\u7c07\u71b5\uff08Cluster Entropy\uff09<\/strong>\uff0c\u52a8\u6001\u8c03\u6574\u89c4\u5212\u5668\u884c\u4e3a\u3002\u8be5\u65b9\u6cd5\u65e0\u9700\u4f9d\u8d56\u9884\u5b9a\u4e49\u7684\u5907\u7528\u5c42\uff08fallback layer\uff09\u6216\u4eba\u5de5\u8bbe\u8ba1\u6210\u672c\u51fd\u6570\uff08expert-crafted cost functions\uff09\uff0c\u76f4\u63a5\u901a\u8fc7\u6570\u636e\u9a71\u52a8\u63d0\u5347\u5b89\u5168\u6027\u548c\u9002\u5e94\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u7c07\u71b5\uff08Cluster Entropy\uff09\u2014\u2014\u65b0\u578b\u4e0d\u786e\u5b9a\u6027\u5ea6\u91cf\u65b9\u6cd5<\/strong><\/p>\n\n\n\n<p>\u9488\u5bf9\u8f68\u8ff9\u8bc4\u5206\uff08trajectory scoring\uff09\u7c7b\u89c4\u5212\u5668\uff0c\u63d0\u51fa\u57fa\u4e8e\u65b9\u5411\u805a\u7c7b\u7684\u65e0\u76d1\u7763\u4e0d\u786e\u5b9a\u6027\u5ea6\u91cf\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5c06\u5019\u9009\u8f68\u8ff9\u6309\u884c\u9a76\u65b9\u5411\uff08\u5982\u201c\u6025\u5de6\u8f6c\u201d\u201c\u7f13\u53f3\u8f6c\u201d\uff09\u805a\u7c7b\u4e3a5\u7c7b\uff0c\u6784\u5efa\u7c7b\u522b\u5206\u5e03\u3002<\/li>\n\n\n\n<li>\u901a\u8fc7\u8ba1\u7b97\u5206\u5e03\u7684\u9999\u519c\u71b5\uff08Shannon Entropy\uff09\uff0c\u91cf\u5316\u89c4\u5212\u51b3\u7b56\u7684\u7f6e\u4fe1\u5ea6\uff0c\u6307\u5bfc\u6d4b\u8bd5\u65f6\u8bad\u7ec3\u7684\u76ee\u6807\u51fd\u6570\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u8be5\u65b9\u6cd5\u7b80\u5355\u3001\u53ef\u89e3\u91ca\uff0c\u4e14\u517c\u5bb9\u4e3b\u6d41\u8f68\u8ff9\u8bc4\u5206\u6a21\u578b\uff08\u5982Hydra-MDP\uff09\u3002<\/p>\n\n\n\n<p><strong>3. \u5b89\u5168\u5173\u952e\u573a\u666f\u57fa\u51c6navsafe\u7684\u6784\u5efa<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u57fa\u4e8eNHTSA\u4e8b\u6545\u7c7b\u578b\u6846\u67b6\uff0c\u4ece\u771f\u5b9e\u6570\u636e\u4e2d\u7b5b\u9009\u5e76\u6807\u6ce810\u7c7b<strong>\u5b89\u5168\u5173\u952e\u573a\u666f<\/strong>\uff08\u5982\u73af\u5c9b\u3001\u65e0\u4fdd\u62a4\u5de6\u8f6c\u3001\u6076\u52a3\u5929\u6c14\u7b49\uff09\uff0c\u5f62\u6210\u7ec6\u7c92\u5ea6\u6d4b\u8bd5\u96c6navsafe\u3002\u8be5\u57fa\u51c6\u63ed\u793a\u4e86\u73b0\u6709\u6a21\u578b\u5728\u8fb9\u7f18\u6848\u4f8b\uff08edge cases\uff09\u4e2d\u7684\u6f5c\u5728\u5931\u6548\u6a21\u5f0f\uff0c\u63a8\u52a8\u81ea\u52a8\u9a7e\u9a76\u5b89\u5168\u6027\u7814\u7a76\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_31173300641_1\"><a href=\"https:\/\/zhida.zhihu.com\/search?content_id=255257139&amp;content_type=Article&amp;match_order=1&amp;q=A+Framework&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">A Framework<\/a><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aA Framework for a Capability-driven Evaluation of Scenario Understanding for Multimodal Large Language Models in Autonomous Driving<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.11400\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.11400<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/v2-79df0491a9f375c51533d49295f8a7e1_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u56db\u7ef4\u80fd\u529b\u8bc4\u4f30\u6846\u67b6<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u51fa\u9996\u4e2a\u7cfb\u7edf\u6027\u8bc4\u4f30MLLMs\u5728\u81ea\u52a8\u9a7e\u9a76\u573a\u666f\u7406\u89e3\u80fd\u529b\u7684\u6846\u67b6\uff0c\u6574\u5408<strong>\u8bed\u4e49\uff08Semantic\uff09\u3001\u7a7a\u95f4\uff08Spatial\uff09\u3001\u65f6\u95f4\uff08Temporal\uff09\u3001\u7269\u7406\uff08Physical\uff09\u56db\u5927\u6838\u5fc3\u7ef4\u5ea6\uff0c\u5e76\u5f15\u5165 \u9884\u6d4b\u80fd\u529b\uff08Anticipation\uff09<\/strong>\u4f5c\u4e3a\u8de8\u7ef4\u5ea6\u5173\u8054\u7684\u5173\u952e\u7ebd\u5e26\uff0c\u7a81\u7834\u4f20\u7edf\u5355\u4e00\u7ef4\u5ea6\u8bc4\u4f30\u7684\u5c40\u9650\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u7ed3\u6784\u5316\u80fd\u529b\u6620\u5c04\u4f53\u7cfb<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u57fa\u4e8e6\u5c42\u573a\u666f\u6a21\u578b\uff086LM\uff09\u3001\u8d23\u4efb\u654f\u611f\u5b89\u5168\u6a21\u578b\uff08RSS\uff09\u53caSafety of the Intended Functionality\uff08SOTIF\uff09\u6807\u51c6\uff0c\u6784\u5efa\u4ece\u73af\u5883\u4e0a\u4e0b\u6587\u2192\u591a\u6a21\u6001\u611f\u77e5\u2192\u573a\u666f\u63cf\u8ff0\u2192\u9884\u6d4b\u63a8\u7406\u2192\u4efb\u52a1\u6267\u884c \u7684\u5b8c\u6574\u6620\u5c04\u94fe\uff0c\u5b9e\u73b0\u5bf9\u52a8\u6001\u4ea4\u901a\u573a\u666f\u7684\u7ec6\u7c92\u5ea6\u89e3\u6790\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u591a\u6a21\u6001\u878d\u5408\u673a\u5236<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u521b\u65b0\u6027\u6574\u5408\u89c6\u89c9\uff08Camera\/LiDAR\uff09\u3001\u8fd0\u52a8\u5b66\uff08IMU\uff09\u3001\u5730\u7406\u7a7a\u95f4\uff08HD Map\uff09\u7b497\u7c7b\u8f93\u5165\u6a21\u6001\uff0c\u901a\u8fc7\u663e\u5f0f\u7a7a\u95f4\u62d3\u6251\u5efa\u6a21 \uff08\u5982\u76f8\u5bf9\u8ddd\u79bb\/\u671d\u5411\/\u5360\u7528\u7a7a\u95f4\uff09\u548c\u9690\u5f0f\u7269\u7406\u7ea6\u675f\u63a8\u7406 \uff08\u5982\u6750\u8d28\u5c5e\u6027\/\u8fd0\u52a8\u5b66\u6a21\u578b\uff09\uff0c\u89e3\u51b3\u4f20\u7edfVLMs\u5728\u7a7a\u95f4-\u65f6\u5e8f\u4fe1\u606f\u8026\u5408\u5904\u7406\u4e0a\u7684\u7f3a\u9677\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>4. \u4efb\u52a1\u9a71\u52a8\u7684\u8bc4\u4f30\u8303\u5f0f<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5b9a\u4e49\u56db\u5927\u4e0b\u6e38\u4efb\u52a1\u7c7b\u578b\uff08\u611f\u77e5\u4efb\u52a1\u03c4_p\u3001\u51b3\u7b56\u4efb\u52a1\u03c4_d\u3001\u4ea4\u4e92\u4efb\u52a1\u03c4_i\u3001\u5b66\u4e60\u4efb\u52a1\u03c4_l\uff09\uff0c\u5e76\u901a\u8fc7<strong>\u94fe\u5f0f\u601d\u7ef4\u63a8\u7406\uff08CoT\uff09<\/strong>\u673a\u5236\u5f3a\u5236\u6a21\u578b\u8f93\u51fa\u7ed3\u6784\u5316\u63cf\u8ff0\uff08\u5982\u5bf9\u8c61\u5c5e\u6027-\u72b6\u6001-\u884c\u4e3a\u9884\u6d4b\uff09\uff0c\u63d0\u5347\u81ea\u52a8\u9a7e\u9a76\u51b3\u7b56\u7684\u53ef\u89e3\u91ca\u6027\u4e0e\u5b89\u5168\u6027\u9a8c\u8bc1\u6548\u7387\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>\u672c\u6587\u5185\u5bb9\u5747\u51fa\u81ea\u300e\u81ea\u52a8\u9a7e\u9a76\u4e4b\u5fc3\u77e5\u8bc6\u661f\u7403\u300f\uff0c\u6b22\u8fce\u52a0\u5165\u4ea4\u6d41\u3002\u8fd9\u91cc\u5df2\u7ecf\u6c47\u805a\u4e86\u8fd14000\u540d\u81ea\u52a8\u9a7e\u9a76\u4ece\u4e1a\u4eba\u5458\uff0c\u6bcf\u65e5\u5206\u4eab\u524d\u6cbf\u6280\u672f\u3001\u884c\u4e1a\u52a8\u6001\u3001\u5c97\u4f4d\u62db\u8058\u3001\u5927\u4f6c\u76f4\u64ad\u7b49\u4e00\u624b\u8d44\u6599\uff01\u6b22\u8fce\u52a0\u5165~<\/strong><\/p>\n\n\n\n<p>https:\/\/t.zsxq.com\/04NneQvBM (\u4e8c\u7ef4\u7801\u81ea\u52a8\u8bc6\u522b)<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_31173300641_2\">DynRsl-VLM<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aDynRsl-VLM: Enhancing Autonomous Driving Perception with Dynamic Resolution Vision-Language Models<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.11265\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.11265<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic1.zhimg.com\/v2-0791970b4158bffe7681dc6b599759fc_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u52a8\u6001\u5206\u8fa8\u7387\u56fe\u50cf\u8f93\u5165\u5904\u7406\u6a21\u5757<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u591a\u5c3a\u5ea6\u7279\u5f81\u4fdd\u7559\u673a\u5236 \uff1a\u901a\u8fc7YOLOv8\u76ee\u6807\u68c0\u6d4b\u751f\u6210\u9ad8\u5206\u8fa8\u7387\u7684\u611f\u5174\u8da3\u533a\u57df\uff08ROIs\uff09\u53ca\u5408\u5e76\u533a\u57df\uff08Combined Regions\uff09\uff0c\u5728\u964d\u4f4e\u5168\u5c40\u56fe\u50cf\u5206\u8fa8\u7387\u7684\u540c\u65f6\u4fdd\u7559\u5173\u952e\u5b9e\u4f53\uff08\u5982\u884c\u4eba\u3001\u8f66\u8f86\uff09\u7684\u7ec6\u8282\u7279\u5f81\u3002<\/li>\n\n\n\n<li>\u8ba1\u7b97\u6548\u7387\u4f18\u5316 \uff1a\u7ed3\u5408\u4e0b\u91c7\u6837\u5168\u5c40\u56fe\u50cf\u4e0e\u9ad8\u5206\u8fa8\u7387\u533a\u57df\u56fe\u50cf\uff0c\u5e73\u8861\u7ec6\u8282\u4fdd\u7559\u4e0e\u8ba1\u7b97\u590d\u6742\u5ea6\uff0c\u89e3\u51b3\u4f20\u7edf\u4e0b\u91c7\u6837\u5bfc\u81f4\u7684\u5c0f\u76ee\u6807\u7279\u5f81\u4e22\u5931\u95ee\u9898\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u52a8\u6001\u5206\u8fa8\u7387\u56fe\u50cf-\u6587\u672c\u5bf9\u9f50\u6a21\u5757<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u591a\u7c92\u5ea6\u7279\u5f81\u4ea4\u4e92 \uff1a\u66ff\u4ee3\u4f20\u7edfQ-Former\uff0c\u901a\u8fc7\u6ce8\u610f\u529b\u673a\u5236\u878d\u5408\u591a\u5206\u8fa8\u7387\u89c6\u89c9\u7279\u5f81\uff08\u5168\u5c40\u3001ROI\u3001\u5408\u5e76\u533a\u57df\uff09\uff0c\u5b9e\u73b0\u8de8\u6a21\u6001\u7ec6\u7c92\u5ea6\u5bf9\u9f50\u3002<\/li>\n\n\n\n<li>\u5bf9\u6bd4\u5b66\u4e60\u589e\u5f3a \uff1a\u91c7\u7528\u5bf9\u79f0InfoNCE\u635f\u5931\u51fd\u6570\uff0c\u7ed3\u5408\u56fe\u50cf-\u6587\u672c\u5bf9\u6bd4\uff08ITC\uff09\u3001\u56fe\u50cf\u5f15\u5bfc\u6587\u672c\u751f\u6210\uff08ITG\uff09\u53ca\u56fe\u6587\u5339\u914d\uff08ITM\uff09\u4efb\u52a1\uff0c\u63d0\u5347\u8de8\u6a21\u6001\u8bed\u4e49\u4e00\u81f4\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u81ea\u52a8\u9a7e\u9a76\u573a\u666f\u9002\u914d\u6027<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u591a\u89c6\u89d2\u878d\u5408 \uff1a\u9002\u914dNuInstruct\u6570\u636e\u96c6\u76846\u89c6\u89d2\u8f93\u5165\uff0c\u901a\u8fc7\u65f6\u7a7a\u9002\u914d\u5668\uff08ST-Adapter\uff09\u6269\u5c55\u57fa\u6a21\u578b\uff08\u5982BLIP-2\uff09\u5904\u7406\u89c6\u9891\u5e8f\u5217\u7684\u80fd\u529b\u3002<\/li>\n\n\n\n<li>\u6027\u80fd\u63d0\u5347\u9a8c\u8bc1 \uff1a\u5728\u611f\u77e5\u3001\u9884\u6d4b\u3001\u98ce\u9669\u8bc4\u4f30\u7b49\u4efb\u52a1\u4e2d\uff0c\u5bf9\u6bd4\u57fa\u7ebf\u6a21\u578b\uff08BLIP-2\u3001BEV-InMLLM\uff09\u5e73\u5747\u63d0\u53473-4%\u7684\u51c6\u786e\u7387\uff0c\u5c24\u5176\u5728\u5c0f\u76ee\u6807\u68c0\u6d4b\u4e0e\u7a7a\u95f4\u5173\u7cfb\u63a8\u7406\u4e2d\u8868\u73b0\u663e\u8457\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_31173300641_3\">DriveGEN<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aDriveGEN: Generalized and Robust 3D Detection in Driving via Controllable Text-to-Image Diffusion Generation<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.11122\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.11122<\/a><\/li>\n\n\n\n<li>\u8bba\u6587\u4ee3\u7801\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/github.com\/Hongbin98\/DriveGEN\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/Hongbin98\/DriveGEN<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-e63c6bab89a9ffabcbb7bf43702ef552_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. \u65e0\u8bad\u7ec3\u53ef\u63a7\u6269\u6563\u751f\u6210\u6846\u67b6<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u9996\u6b21\u63d0\u51fa\u8bad\u7ec3-free\u7684\u53ef\u63a7\u6587\u672c-\u56fe\u50cf\uff08T2I\uff09\u6269\u6563\u751f\u6210\u65b9\u6cd5 \uff0c\u76f4\u63a5\u5229\u7528\u9884\u8bad\u7ec3\u6269\u6563\u6a21\u578b\uff08\u5982Stable Diffusion\uff09\u751f\u6210\u591a\u6837\u5316OOD\u9a7e\u9a76\u573a\u666f\uff08\u5982\u96ea\u5929\u3001\u96fe\u5929\uff09\uff0c\u65e0\u9700\u989d\u5916\u8bad\u7ec3\u6269\u6563\u6a21\u578b\uff0c\u663e\u8457\u964d\u4f4e\u6d4b\u8bd5\u65f6\u8ba1\u7b97\u6210\u672c\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u53cc\u9636\u6bb5\u51e0\u4f55\u4fdd\u771f\u589e\u5f3a\u673a\u5236<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u81ea\u539f\u578b\u63d0\u53d6\uff08Self-Prototype Extraction\uff09<\/li>\n<\/ul>\n\n\n\n<p>\u901a\u8fc7\u5e03\u5c40\uff08bounding box\uff09\u5f15\u5bfc\u7684\u5cf0\u503c\u51fd\u6570\u91cd\u52a0\u6743 \uff0c\u4ece\u81ea\u6ce8\u610f\u529b\u7279\u5f81\u4e2d\u63d0\u53d6\u7ec6\u7c92\u5ea6\u81ea\u539f\u578b\uff08Self-Prototypes\uff09 \uff0c\u7cbe\u51c6\u5b9a\u4f4d\u591a\u7269\u4f53\u533a\u57df\uff0c\u89e3\u51b3\u4f20\u7edf\u65b9\u6cd5\u56e0\u4f9d\u8d56\u7c97\u7c92\u5ea6\u81ea\u6ce8\u610f\u529b\u5bfc\u81f4\u7684\u7269\u4f53\u9057\u6f0f\/\u9519\u4f4d\u95ee\u9898\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u539f\u578b\u5f15\u5bfc\u6269\u6563\uff08Prototype-Guided Diffusion\uff09<\/li>\n\n\n\n<li>\u8bed\u4e49\u611f\u77e5\u7279\u5f81\u5bf9\u9f50\uff08Semantic-aware Alignment\uff09 \uff1a\u5728\u53bb\u566a\u9636\u6bb5\u5bf9\u9f50\u81ea\u539f\u578b\u4e0e\u81ea\u6ce8\u610f\u529b\u7279\u5f81\uff0c\u4fdd\u7559\u7269\u4f53\u8bed\u4e49\u7ed3\u6784\uff1b<\/li>\n\n\n\n<li>\u6d45\u5c42\u7279\u5f81\u5bf9\u9f50\uff08Shallow Feature Alignment\uff09 \uff1a\u901a\u8fc7\u7ea6\u675f\u6d45\u5c42\u566a\u58f0\u6f5c\u5728\u7279\u5f81\uff0c\u8865\u507f\u8bed\u4e49\u5bf9\u9f50\u5728\u5c0f\u7269\u4f53\u51e0\u4f55\u7ec6\u8282\u4e0a\u7684\u4e0d\u8db3\uff0c\u786e\u4fdd\u5fae\u5c0f\u7269\u4f53\uff08\u5982\u9a91\u884c\u8005\uff09\u76843D\u51e0\u4f55\u7cbe\u5ea6\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u8de8\u573a\u666f\u6cdb\u5316\u6027\u63d0\u5347<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728KITTI\u548cnuScenes\u6570\u636e\u96c6\u9a8c\u8bc1\uff0c\u4ec5\u9700\u5355\u6b21\u589e\u5f3a\uff08\u5982\u751f\u6210&#8221;\u96ea\u5929&#8221;\u573a\u666f\uff09\uff0c\u5373\u53ef\u4f7f3D\u68c0\u6d4b\u5668\u572813\u79cdOOD\u573a\u666f\uff08\u566a\u58f0\u3001\u6a21\u7cca\u3001\u5929\u6c14\u53d8\u5316\u7b49\uff09\u5e73\u5747mAP\u63d0\u53477.6%\uff0c\u4e14\u5728\u672a\u89c1\u8fc7\u7684&#8221;\u96fe\u5929&#8221;&#8221;\u6563\u7126&#8221;\u573a\u666f\u4e2d\u4ea6\u6709\u663e\u8457\u6cdb\u5316\u80fd\u529b\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_31173300641_4\"><a href=\"https:\/\/zhida.zhihu.com\/search?content_id=255257139&amp;content_type=Article&amp;match_order=1&amp;q=AugMapNet&amp;zhida_source=entity\" target=\"_blank\" rel=\"noreferrer noopener\">AugMapNet<\/a><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aAugMapNet: Improving Spatial Latent Structure via BEV Grid Augmentation for Enhanced Vectorized Online HD Map Construction<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.13430\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.13430<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pica.zhimg.com\/v2-3dea4e1457b312defc43eb4762a75f1c_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/strong><\/p>\n\n\n\n<p><strong>1. BEV\u7f51\u683c\u589e\u5f3a\uff08BEV Grid Augmentation\uff09<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u51fa\u4e00\u79cd\u65b0\u578b\u6f5c\u5728\u7a7a\u95f4\u589e\u5f3a\u673a\u5236\uff0c\u901a\u8fc7\u5c06\u5149\u6805\u89e3\u7801\u5668\u9884\u6d4b\u7684\u4e2d\u95f4\u5149\u6805\u5730\u56fe\uff08Raster Map\uff09\u7f16\u7801\u4e3a\u7a7a\u95f4\u7279\u5f81\uff0c\u5e76\u4e0e\u539f\u59cbBEV\u6f5c\u5728\u7f51\u683c\u76f8\u52a0\uff0c\u76f4\u63a5\u6ce8\u5165\u5bc6\u96c6\u7a7a\u95f4\u76d1\u7763\u4fe1\u606f\uff0c\u4ece\u800c\u5f3a\u5316BEV\u6f5c\u5728\u7a7a\u95f4\u7684\u7ed3\u6784\u5316\u8868\u5f81\u80fd\u529b\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u68af\u5ea6\u9694\u79bb\u7b56\u7565\uff08Gradient Stopping\uff09<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5f15\u5165\u68af\u5ea6\u505c\u6b62\u6280\u672f\uff0c\u963b\u65ad\u5149\u6805\u89e3\u7801\u5206\u652f\u7684\u68af\u5ea6\u53cd\u5411\u4f20\u64ad\u81f3BEV\u7f16\u7801\u5668\uff0c\u907f\u514d\u5149\u6805\u76d1\u7763\u4efb\u52a1\u5e72\u6270\u77e2\u91cf\u89e3\u7801\u5668\u7684\u6f5c\u5728\u7a7a\u95f4\u4f18\u5316\uff0c\u786e\u4fdd\u77e2\u91cf\u89e3\u7801\u7684\u7a33\u5b9a\u6027\u548c\u72ec\u7acb\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u53cc\u9636\u6bb5BEV\u7279\u5f81\u4f18\u5316<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bbe\u8ba1\u4e24\u7ea7\u5377\u79ef\u7f51\u7edc\uff08CNN1\u548cCNN2\uff09\uff0c\u5206\u522b\u5bf9\u539f\u59cbBEV\u7279\u5f81\u8fdb\u884c\u5bf9\u9f50\uff08Align\uff09\u548c\u589e\u5f3a\u540e\u7279\u5f81\u878d\u5408\uff0c\u8fdb\u4e00\u6b65\u4f18\u5316\u6f5c\u5728\u7a7a\u95f4\u7684\u7ed3\u6784\u5316\u7279\u6027\uff0c\u63d0\u5347\u957f\u8ddd\u79bb\u611f\u77e5\u4e0e\u590d\u6742\u573a\u666f\uff08\u5982\u4eba\u884c\u6a2a\u9053\uff09\u7684\u68c0\u6d4b\u7cbe\u5ea6\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>4. \u5bc6\u96c6\u4e0e\u7a00\u758f\u76d1\u7763\u7684\u534f\u540c\u4f18\u5316<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u901a\u8fc7\u5e76\u884c\u5149\u6805\u89e3\u7801\u4efb\u52a1\u63d0\u4f9b\u5bc6\u96c6\u50cf\u7d20\u7ea7\u76d1\u7763\uff0c\u540c\u65f6\u7ed3\u5408\u77e2\u91cf\u89e3\u7801\u7684\u5b9e\u4f8b\u7ea7\u7a00\u758f\u76d1\u7763\uff0c\u5728\u65e0\u9700\u590d\u6742\u540e\u5904\u7406\u7684\u60c5\u51b5\u4e0b\u5b9e\u73b0\u9ad8\u6548\u77e2\u91cf\u5316\u5730\u56fe\u751f\u6210\uff0c\u663e\u8457\u63d0\u5347\u6a21\u578b\u5bf9\u52a8\u6001\u73af\u5883\uff08\u5982\u65bd\u5de5\u533a\u57df\u3001\u5929\u6c14\u53d8\u5316\uff09\u7684\u9c81\u68d2\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>5. \u6f5c\u5728\u7a7a\u95f4\u7ed3\u6784\u9a8c\u8bc1<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u91c7\u7528Silhouette\u8bc4\u5206\u3001\u4e3b\u6210\u5206\u5206\u6790\uff08PCA\uff09\u548c\u4e92\u4fe1\u606f\uff08Mutual Information\uff09\u91cf\u5316\u6f5c\u5728\u7a7a\u95f4\u7684\u7ed3\u6784\u5316\u7a0b\u5ea6\uff0c\u8bc1\u660e\u589e\u5f3a\u540e\u7684BEV\u7f51\u683c\u5177\u6709\u66f4\u9ad8\u7684\u7c7b\u5185\u805a\u7c7b\u6027\u548c\u7c7b\u95f4\u533a\u5206\u5ea6\uff0c\u4e14\u4e0e\u771f\u5b9e\u5149\u6805\u5730\u56fe\u7684\u8bed\u4e49\u4e00\u81f4\u6027\u663e\u8457\u63d0\u5347\uff0c\u76f4\u63a5\u5173\u8054\u77e2\u91cf\u5316\u9884\u6d4b\u6027\u80fd\u7684\u6539\u8fdb\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h_31173300641_5\">SparseAlign<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aSparseAlign: A Fully Sparse Framework for Cooperative Object Detection<\/li>\n\n\n\n<li>\u8bba\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/link.zhihu.com\/?target=https%3A\/\/arxiv.org\/abs\/2503.12982\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/arxiv.org\/abs\/2503.12982<\/a><\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/pic3.zhimg.com\/v2-452cff62bb0e79da80114233d4fd8ef6_1440w.jpg\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u6838\u5fc3\u521b\u65b0\u70b9\uff1a<\/p>\n\n\n\n<p><strong>1. \u5168\u7a00\u758f\u67b6\u6784\u8bbe\u8ba1<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u51faSUNet\u7a00\u758f3D\u4e3b\u5e72\u7f51\u7edc\uff0c\u901a\u8fc7<strong>\u5750\u6807\u6269\u5c55\u5377\u79ef\uff08CEC\uff09<\/strong>\u89e3\u51b3\u7a00\u758f\u5377\u79ef\u7684\u4e24\u5927\u56fa\u6709\u95ee\u9898\uff1a<\/li>\n\n\n\n<li><strong>\u4e2d\u5fc3\u7279\u5f81\u7f3a\u5931\uff08CFM\uff09<\/strong> \uff1a\u901a\u8fc72D CEC\u6269\u5c55BEV\u5750\u6807\u8986\u76d6\u7269\u4f53\u4e2d\u5fc3\u533a\u57df<\/li>\n\n\n\n<li><strong>\u5b64\u7acb\u5377\u79ef\u57df\uff08ICF\uff09<\/strong> \uff1a\u901a\u8fc73D CEC\u589e\u5f3a\u8fdc\u8ddd\u79bb\/\u906e\u6321\u533a\u57df\u7684\u70b9\u4e91\u8fde\u901a\u6027<\/li>\n<\/ul>\n\n\n\n<p><strong>2. \u65f6\u7a7a\u5bf9\u9f50\u6a21\u5757<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>TempAlign\u6a21\u5757 \uff1a\u57fa\u4e8e\u67e5\u8be2\u7684\u65f6\u5e8f\u4e0a\u4e0b\u6587\u5b66\u4e60\uff0c\u901a\u8fc7\u5386\u53f2\u5e27\u7279\u5f81\u8865\u507f\u901a\u4fe1\u5ef6\u8fdf\u548c\u4f20\u611f\u5668\u5f02\u6b65<\/li>\n\n\n\n<li>PoseAlign\u6a21\u5757 \uff1a\u63d0\u51fa\u59ff\u6001\u65e0\u5173\u90bb\u57df\u56fe\u7279\u5f81\u5339\u914d \uff0c\u7ed3\u5408PGO\u4f18\u5316\u5b9e\u73b0\u9c81\u68d2\u7684\u76f8\u5bf9\u4f4d\u59ff\u6821\u6b63<\/li>\n\n\n\n<li>SpatialAlign\u6a21\u5757 \uff1a\u8bbe\u8ba1\u5750\u6807\u81ea\u9002\u5e94\u878d\u5408\u673a\u5236 \uff0c\u901a\u8fc7MLP\u53d8\u6362\u4e0e\u6700\u8fd1\u90bb\u7279\u5f81\u805a\u5408\u5b9e\u73b0\u591a\u667a\u80fd\u4f53\u7279\u5f81\u5bf9\u9f50<\/li>\n<\/ul>\n\n\n\n<p><strong>3. \u9ad8\u6548\u901a\u4fe1\u673a\u5236<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u91c7\u7528\u5bf9\u8c61\u67e5\u8be2\u7279\u5f81\u5171\u4eab \uff0c\u76f8\u6bd4\u4f20\u7edfBEV\u7279\u5f81\u56fe\u51cf\u5c1198%\u5e26\u5bbd\u6d88\u8017<\/li>\n\n\n\n<li>\u63d0\u51faCompassRose\u7f16\u7801 \u4f18\u53163D\u6846\u65b9\u5411\u56de\u5f52\uff0c\u901a\u8fc7\u56db\u951a\u70b9\u89d2\u5ea6\u5d4c\u5165\u63d0\u5347\u65b9\u5411\u9884\u6d4b\u9c81\u68d2\u6027<\/li>\n<\/ul>\n\n\n\n<p><strong>4. \u6027\u80fd\u7a81\u7834<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728OPV2V\/DairV2X\u7b49\u6570\u636e\u96c6\u4e0a\uff0c\u4ee5\u5168\u7a00\u758f\u67b6\u6784\u8d85\u8d8a\u73b0\u6709\u5bc6\u96c6\u65b9\u6cd5\uff0c\u5728TA-COOD\u4efb\u52a1\u4e2dAP@0.7 \u6307\u6807\u63d0\u5347\u8fbe14.4%<\/li>\n\n\n\n<li>\u901a\u8fc7\u81ea\u7531\u7a7a\u95f4\u589e\u5f3a\uff08FSA\uff09\u548c \u68af\u5ea6\u8ba1\u7b97\u8c03\u5ea6 \u7b56\u7565\uff0c\u663e\u8457\u63d0\u5347\u8fdc\u8ddd\u79bb\u68c0\u6d4b\u548c\u8bad\u7ec3\u6548\u7387<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u539f\u6587\u94fe\u63a5\uff1ahttps:\/\/zhuanlan.z [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3,1],"tags":[],"class_list":["post-26486","post","type-post","status-publish","format-standard","hentry","category-technology-frontier","category-home"],"views":0,"_links":{"self":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/26486"}],"collection":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/comments?post=26486"}],"version-history":[{"count":1,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/26486\/revisions"}],"predecessor-version":[{"id":26487,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/posts\/26486\/revisions\/26487"}],"wp:attachment":[{"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/media?parent=26486"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/categories?post=26486"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/222.128.65.89:18086\/index.php\/wp-json\/wp\/v2\/tags?post=26486"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}