{"id":12689,"date":"2026-09-29T04:20:40","date_gmt":"2026-09-28T20:20:40","guid":{"rendered":"https:\/\/infernews.com\/blog\/lightmis-ultra-lightweight-medical-image-segmentation-without-a-stage-wise-decoder\/"},"modified":"2026-09-29T04:20:40","modified_gmt":"2026-09-28T20:20:40","slug":"lightmis-ultra-lightweight-medical-image-segmentation-without-a-stage-wise-decoder","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/lightmis-ultra-lightweight-medical-image-segmentation-without-a-stage-wise-decoder\/","title":{"rendered":"LightMIS \u4ee5\u8f15\u91cf\u5316\u67b6\u69cb\u63a8\u9032\u91ab\u5b78\u5f71\u50cf\u5206\u5272"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/09\/pasted-c3f65218bfa9.jpg\" alt=\"LightMIS accuracy\u2013latency Pareto front\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728\u624b\u6a5f GPU \u4e0a\u57f7\u884c\u91ab\u5b78\u5f71\u50cf\u5206\u5272\uff0c\u6a21\u578b\u5927\u5c0f\u8207\u8655\u7406\u901f\u5ea6\u90fd\u6703\u5f71\u97ff\u90e8\u7f72\u9078\u64c7\u3002LightMIS \u662f\u4e00\u6b3e\u9762\u5411 2D \u4e8c\u5143\u91ab\u5b78\u5f71\u50cf\u5206\u5272\u7684\u8f15\u91cf\u6a21\u578b\uff0c\u900f\u904e\u7cbe\u7c21\u591a\u5c3a\u5ea6\u7de8\u78bc\u5668\u548c\u805a\u5408\u5f0f\u8f38\u51fa\u982d\uff0c\u6e1b\u5c11\u50b3\u7d71\u9010\u968e\u89e3\u78bc\u5668\u5e36\u4f86\u7684\u91cd\u8907\u4e0a\u63a1\u6a23\u53ca\u7279\u5fb5\u878d\u5408\u904b\u7b97\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6a21\u578b\u5728\u6bcf\u500b\u7a7a\u9593\u5c3a\u5ea6\u4f7f\u7528 Adaptive Fusion Cascade\uff08AFC\uff09\u8655\u7406\u7279\u5fb5\uff1b\u5176\u4e2d\u7684 Progressive Receptive Fusion\uff08PRF\uff09\u6703\u66ab\u6642\u64f4\u5c55\u901a\u9053\uff0c\u914d\u5408\u4e0d\u540c\u6df1\u5ea6\u5377\u7a4d\u611f\u53d7\u91ce\u53ca\u5206\u652f\u9593\u9010\u6b65\u50b3\u905e\u8cc7\u8a0a\uff0c\u88dc\u8db3\u7a84\u901a\u9053\u7279\u5fb5\u7684\u8868\u9054\u80fd\u529b\u3002\u7de8\u78bc\u5668\u5404\u5c64\u8f38\u51fa\u518d\u7531 Scale-Aligned Projection\uff08SAP\uff09\u5c0d\u9f4a\u81f3\u76f8\u540c\u89e3\u6790\u5ea6\u8207\u901a\u9053\u5bec\u5ea6\uff0c\u878d\u5408\u5f8c\u4ea4\u7531\u6700\u5f8c\u4e00\u500b AFC refinement\uff0c\u6bcb\u9808\u6cbf\u7528\u9010\u5c64\u4e0a\u63a1\u6a23\u7684\u89e3\u78bc\u7d50\u69cb\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728 nnU-Net v2.3.1 \u7684\u5171\u540c\u6e2c\u8a66\u6d41\u7a0b\u4e0b\uff0c\u7814\u7a76\u4ee5\u516d\u500b\u8cc7\u6599\u96c6\u3001\u4e94\u500b\u5f71\u50cf\u9818\u57df\u53ca\u4e94\u6298\u4ea4\u53c9\u9a57\u8b49\u8a55\u4f30\u6a21\u578b\u3002LightMIS \u7684\u5340\u57df\u91cd\u758a\u6307\u6a19\u8207\u660e\u986f\u66f4\u5927\u7684 Mobile U-ViT\u3001nnWNet \u548c nnU-Net \u63a5\u8fd1\uff0c\u908a\u754c\u8ddd\u96e2\u7d50\u679c\u4ea6\u5177\u7af6\u722d\u529b\uff1b\u53c3\u6578\u91cf\u5c11 90.58% \u81f3 99.61%\uff0cGFLOPs \u5c11 82.54% \u81f3 96.14%\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728\u914d\u5099 Arm Mali-G52 MC2 GPU \u7684\u624b\u6a5f\u4e0a\uff0c\u5404\u7248\u672c\u90fd\u9054\u5230 100% GPU \u7b97\u5b50\u8986\u84cb\u7387\uff1bLightMIS-T \u7684\u59d4\u6d3e\u4e2d\u4f4d\u5ef6\u9072\u70ba 53.31 \u6beb\u79d2\uff0cLightMIS \u70ba 138.31 \u6beb\u79d2\u3002\u9019\u4e9b\u7d50\u679c\u9069\u5408\u53c3\u8003\u65bc\u9700\u8981\u63a7\u5236\u6a21\u578b\u9ad4\u7a4d\u3001\u904b\u7b97\u91cf\u6216\u624b\u6a5f\u7aef\u63a8\u7406\u7684\u7814\u7a76\u8207\u958b\u767c\u5de5\u4f5c\uff0c\u4f46\u4e0d\u4ee3\u8868\u6240\u6709\u88dd\u7f6e\u548c\u91ab\u7642\u5834\u666f\u90fd\u6703\u6709\u76f8\u540c\u901f\u5ea6\u6216\u6e96\u78ba\u5ea6\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8981\u91cd\u73fe\u6e2c\u8a66\uff0c\u9700\u914d\u7f6e nnU-Net v2.3.1\uff0c\u5c07 DRIVE\u3001Kvasir-SEG\u3001DSB18\u3001BUSI\u3001ISIC-2017 \u6216 ISIC-2018 \u7b49\u8cc7\u6599\u96c6\u8f49\u6210 nnU-Net \u683c\u5f0f\uff0c\u518d\u57f7\u884c 2D \u898f\u5283\u3001\u9810\u8655\u7406\u53ca\u4e94\u6298\u8a13\u7df4\u3002\u5132\u5b58\u5eab\u63d0\u4f9b\u8cc7\u6599\u8f49\u63db\u8173\u672c\u548c\u8a13\u7df4\u5668\u6a94\u6848\uff1b\u6587\u7ae0\u8349\u7a3f\u6240\u64da\u5167\u5bb9\u672a\u5305\u542b\u5b8c\u6574\u547d\u4ee4\u53ca\u786c\u4ef6\u90e8\u7f72\u6d41\u7a0b\uff0c\u4e0d\u80fd\u64da\u6b64\u63a8\u65b7\u5b89\u88dd\u5f8c\u5373\u53ef\u76f4\u63a5\u5728\u624b\u6a5f\u57f7\u884c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u9069\u7528\u4efb\u52d9\uff1a<\/strong> 2D \u4e8c\u5143\u91ab\u5b78\u5f71\u50cf\u5206\u5272\u3002<\/li>\n<li><strong>\u6838\u5fc3\u505a\u6cd5\uff1a<\/strong> AFC \u8207 PRF\u5f37\u5316\u591a\u5c3a\u5ea6\u7279\u5fb5\uff0cSAP \u5c0d\u9f4a\u4e26\u805a\u5408\u7de8\u78bc\u5668\u8f38\u51fa\u3002<\/li>\n<li><strong>\u4e3b\u8981\u53d6\u6368\uff1a<\/strong> \u6e1b\u5c11\u53c3\u6578\u91cf\u8207 GFLOPs\uff0c\u4e26\u7701\u53bb\u50b3\u7d71\u9010\u968e\u89e3\u78bc\u5668\u3002<\/li>\n<li><strong>\u8a55\u4f30\u7bc4\u570d\uff1a<\/strong> \u516d\u500b\u8cc7\u6599\u96c6\u3001\u4e94\u500b\u5f71\u50cf\u9818\u57df\uff0c\u63a1\u7528\u4e94\u6298\u4ea4\u53c9\u9a57\u8b49\u3002<\/li>\n<li><strong>\u90e8\u7f72\u53c3\u8003\uff1a<\/strong> \u624b\u6a5f Arm Mali-G52 MC2 \u4e0a\u5404\u7248\u672c\u9054\u5230\u5b8c\u6574 GPU \u7b97\u5b50\u8986\u84cb\uff0c\u4f46\u5ef6\u9072\u56e0\u7248\u672c\u800c\u7570\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/huggingface.co\/papers\/2609.28327\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/AndreiiArhire\/LightMIS\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>LightMIS\u4ee5\u7cbe\u7c21\u591a\u5c3a\u5ea6\u7de8\u78bc\u5668\u53d6\u4ee3\u50b3\u7d71\u9010\u968e\u89e3\u78bc\u5668\uff0c\u5617\u8a66\u5728\u6e1b\u5c11\u6a21\u578b\u53c3\u6578\u8207\u904b\u7b97\u91cf\u7684\u540c\u6642\uff0c\u7dad\u6301\u91ab\u5b78\u5f71\u50cf\u5206\u5272\u8868\u73fe\u3002<\/p>\n","protected":false},"author":8,"featured_media":12688,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,76,127,30,144,150],"tags":[],"class_list":["post-12689","post","type-post","status-publish","format-standard","hentry","category-133","category-76","category-127","category-image","category-medical","category-150"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12689","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=12689"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12689\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/12688"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=12689"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=12689"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=12689"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}