{"id":21162,"date":"2024-07-15T10:14:18","date_gmt":"2024-07-15T02:14:18","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=21162"},"modified":"2024-07-15T10:16:18","modified_gmt":"2024-07-15T02:16:18","slug":"meta%e5%85%ac%e5%8f%b8%e5%bc%80%e6%ba%90%e5%a4%a7%e6%95%b0%e6%8d%ae%e6%a8%a1%e5%9e%8bsam%e5%ae%9e%e6%88%98%e6%bc%94%e7%bb%83","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2024\/07\/15\/21162\/","title":{"rendered":"Meta\u516c\u53f8\u5f00\u6e90\u5927\u6570\u636e\u6a21\u578bSAM\u5b9e\u6218\u6f14\u7ec3"},"content":{"rendered":"<p style=\"font-weight: 400;\">\u672c\u6587\u9996\u5148\u4ecb\u7ecdMeta\u516c\u53f8\u5f00\u53d1\u7684\u5f00\u6e90\u56fe\u50cf\u5206\u5272\u6a21\u578bSAM\u7684\u67b6\u6784\uff0c\u7136\u540e\u901a\u8fc7\u4e00\u4e2a\u6cb3\u6d41\u50cf\u7d20\u5206\u5272\u9065\u611f\u4efb\u52a1\u7684\u5b9e\u6218\u6848\u4f8b\u5c55\u793aSAM\u6a21\u578b\u5e94\u7528\u5f00\u53d1\u6d89\u53ca\u7684\u5173\u952e\u6280\u672f\u4e0e\u6a21\u578b\u4f18\u52bf\u3002<\/p>\n<p><img data-dominant-color=\"7f8484\" data-has-transparency=\"false\" style=\"--dominant-color: #7f8484;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-21166\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/07\/9361c102031b64cb7ec464a06289604b358692-1-300x199-1.png\" width=\"300\" height=\"199\" alt=\"\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/07\/9361c102031b64cb7ec464a06289604b358692-1-300x199-1.png 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/07\/9361c102031b64cb7ec464a06289604b358692-1-300x199-1-150x100.png 150w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/p>\n<p>\u5f53\u524d\uff0c\u8bb8\u591a\u5f3a\u5927\u7684\u5f00\u6e90\u57fa\u7840\u6a21\u578b\u7684\u53d1\u5e03\uff0c\u52a0\u4e0a\u5fae\u8c03\u6280\u672f\u7684\u4e0d\u65ad\u8fdb\u6b65\uff0c\u5df2\u7ecf\u5e26\u6765\u4e86\u673a\u5668\u5b66\u4e60\u548c\u4eba\u5de5\u667a\u80fd\u7684\u65b0\u8303\u5f0f\u3002\u6574\u4f53\u6765\u770b\uff0c\u8fd9\u573a\u9769\u547d\u7684\u6838\u5fc3\u5728\u4e8e\u8f6c\u6362\u5668\u6a21\u578b\uff08https:\/\/arxiv.org\/pdf\/1706.03762\uff09\u3002<\/p>\n<p>\u867d\u7136\u9664\u4e86\u8d44\u91d1\u5145\u8db3\u7684\u516c\u53f8\u4e4b\u5916\uff0c\u6240\u6709\u516c\u53f8\u90fd\u66fe\u7ecf\u65e0\u6cd5\u83b7\u5f97\u9ad8\u7cbe\u5ea6\u7684\u7279\u5b9a\u9886\u57df\u6a21\u578b\uff1b\u4f46\u662f\u5982\u4eca\uff0c\u57fa\u7840\u6a21\u578b\u8303\u5f0f\u751a\u81f3\u80fd\u591f\u5141\u8bb8\u5b66\u751f\u6216\u72ec\u7acb\u7814\u7a76\u4eba\u5458\u83b7\u5f97\u9002\u5ea6\u7684\u8d44\u6e90\uff0c\u4ee5\u5b9e\u73b0\u4e0e\u6700\u5148\u8fdb\u7684\u4e13\u6709\u6a21\u578b\u76f8\u6297\u8861\u7684\u7ed3\u679c\u3002<\/p>\n<p>\u5fae\u8c03\u6280\u672f\u53ef\u4ee5\u6781\u5927\u5730\u63d0\u9ad8\u6a21\u578b\u4efb\u52a1\u7684\u6027\u80fd<\/p>\n<p>\u672c\u6587\u65e8\u5728\u63a2\u8ba8Meta\u516c\u53f8\u7684\u5206\u5272\u4e00\u5207\u6a21\u578b\uff08SAM\uff1aSegment Anything Model\uff09\u5728\u6cb3\u6d41\u50cf\u7d20\u5206\u5272\u9065\u611f\u4efb\u52a1\u4e2d\u7684\u5e94\u7528\u3002\u5982\u679c\u60a8\u60f3\u76f4\u63a5\u8df3\u5230\u6709\u5173\u9879\u76ee\u4ee3\u7801\uff0c\u90a3\u4e48\u8fd9\u4e2a\u9879\u76ee\u7684\u6e90\u6587\u4ef6\u53ef\u4ee5\u5728GitHub\uff08https:\/\/github.com\/geo-smart\/water-surf\/blob\/main\/book\/chapters\/masking_distributed.ipynb\uff09\u4e0a\u83b7\u5f97\uff0c\u6570\u636e\u4e5f\u53ef\u4ee5\u5728HuggingFace\uff08https:\/\/huggingface.co\/datasets\/stodoran\/elwha-segmentation-v1\uff09\u4e0a\u627e\u5230\u3002\u5f53\u7136\uff0c\u6211\u8fd8\u662f\u5efa\u8bae\u60a8\u5148\u9605\u8bfb\u4e00\u4e0b\u672c\u6587\u3002<\/p>\n<h3>\u9879\u76ee\u9700\u6c42<\/h3>\n<p>\u7b2c\u4e00\u4e2a\u4efb\u52a1\u662f\u627e\u5230\u6216\u521b\u5efa\u4e00\u4e2a\u5408\u9002\u7684\u6570\u636e\u96c6\u3002\u6839\u636e\u73b0\u6709\u6587\u732e\uff0c\u4e00\u4e2a\u826f\u597d\u7684SAM\u5fae\u8c03\u6570\u636e\u96c6\u5c06\u81f3\u5c11\u5305\u542b200\u2013800\u5f20\u56fe\u50cf\u3002\u8fc7\u53bb\u5341\u5e74\u6df1\u5ea6\u5b66\u4e60\u8fdb\u6b65\u7684\u4e00\u4e2a\u5173\u952e\u6559\u8bad\u662f\uff0c\u6570\u636e\u8d8a\u591a\u8d8a\u597d\uff0c\u8fd9\u6837\u4e00\u6765\uff0c\u66f4\u5927\u89c4\u6a21\u7684\u5fae\u8c03\u6570\u636e\u96c6\u5c31\u4e0d\u4f1a\u51fa\u9519\u3002\u7136\u800c\uff0c\u57fa\u7840\u6a21\u578b\u7814\u7a76\u7684\u4e00\u4e2a\u91cd\u8981\u76ee\u6807\u662f\uff0c\u5141\u8bb8\u5373\u4f7f\u662f\u76f8\u5bf9\u8f83\u5c0f\u7684\u6570\u636e\u96c6\u4e5f\u8db3\u4ee5\u5b9e\u73b0\u5f3a\u5927\u7684\u6027\u80fd\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u6211\u4eec\u8fd8\u9700\u8981\u6709\u4e00\u4e2aHuggingFace\u5e10\u6237\uff0c\u8fd9\u53ef\u4ee5\u5728\u94fe\u63a5https:\/\/huggingface.co\/join\u5904\u521b\u5efa\u3002\u4f7f\u7528HuggingFace\uff0c\u6211\u4eec\u53ef\u4ee5\u968f\u65f6\u4ece\u4efb\u4f55\u8bbe\u5907\u8f7b\u677e\u5b58\u50a8\u548c\u83b7\u53d6\u6570\u636e\u96c6\uff0c\u8fd9\u4f7f\u5f97\u534f\u4f5c\u548c\u518d\u73b0\u6027\u66f4\u52a0\u5bb9\u6613\u3002<\/p>\n<p>\u6700\u540e\u4e00\u4e2a\u9700\u6c42\u662f\uff0c\u5177\u5907\u4e00\u53f0\u5e26\u6709GPU\u7684\u8bbe\u5907\uff0c\u6211\u4eec\u53ef\u4ee5\u5728\u5176\u4e0a\u8fd0\u884c\u8bad\u7ec3\u5de5\u4f5c\u6d41\u7a0b\u3002\u901a\u8fc7Google Colab\uff08https:\/\/colab.research.google.com\/\uff09\u514d\u8d39\u63d0\u4f9b\u7684Nvidia T4 GPU\u8db3\u591f\u5f3a\u5927\uff0c\u53ef\u4ee5\u572812\u5c0f\u65f6\u5185\u5bf91000\u5f20\u56fe\u50cf\u8fdb\u884c50\u4e2a\u65f6\u671f\u7684\u6700\u5927SAM\u6a21\u578b\u68c0\u67e5\u70b9\uff08sam-vit-huge\uff09\u8bad\u7ec3\u3002<\/p>\n<p>\u4e3a\u4e86\u907f\u514d\u5728\u6258\u7ba1\u8fd0\u884c\u65f6\u56e0\u4f7f\u7528\u9650\u5236\u800c\u5f71\u54cd\u8fdb\u5ea6\uff0c\u60a8\u53ef\u4ee5\u5b89\u88c5Google Drive\u5e76\u5c06\u6bcf\u4e2a\u6a21\u578b\u68c0\u67e5\u70b9\u4fdd\u5b58\u5728\u90a3\u91cc\u3002\u6216\u8005\uff0c\u90e8\u7f72\u5e76\u8fde\u63a5\u5230GCP\u865a\u62df\u673a\uff08https:\/\/console.cloud.google.com\/marketplace\/product\/colab-marketplace-image-public\/colab\uff09\u4ee5\u5b8c\u5168\u7ed5\u8fc7\u9650\u5236\u3002\u5982\u679c\u60a8\u4ee5\u524d\u4ece\u672a\u4f7f\u7528\u8fc7GCP\uff0c\u90a3\u4e48\u60a8\u5c31\u6709\u8d44\u683c\u83b7\u5f97300\u7f8e\u5143\u7684\u514d\u8d39\u4fe1\u8d37\uff0c\u8fd9\u8db3\u4ee5\u652f\u6301\u5bf9\u6a21\u578b\u8fdb\u884c\u81f3\u5c11\u5341\u51e0\u6b21\u7684\u8bad\u7ec3\u4e86\u3002<\/p>\n<h3>\u7406\u89e3SAM\u67b6\u6784<\/h3>\n<p>\u5728\u5f00\u59cb\u8bad\u7ec3\u4e4b\u524d\uff0c\u6211\u4eec\u9700\u8981\u5148\u6765\u4e86\u89e3\u4e00\u4e0bSAM\u6a21\u578b\u7684\u67b6\u6784\u3002\u8be5\u6a21\u578b\u5305\u542b\u4e09\u4e2a\u7ec4\u4ef6\uff1a\u4e00\u4e2a\u662f\u4ece\u7a0d\u7ecf\u4fee\u6539\u7684\u63a9\u7801\u81ea\u52a8\u7f16\u7801\u5668\uff08https:\/\/arxiv.org\/pdf\/2111.06377\uff09\u5f97\u5230\u7684\u56fe\u50cf\u7f16\u7801\u5668\uff0c\u4e00\u4e2a\u80fd\u591f\u5904\u7406\u5404\u79cd\u63d0\u793a\u7c7b\u578b\u7684\u76f8\u5f53\u7075\u6d3b\u7684\u63d0\u793a\u7f16\u7801\u5668\uff0c\u8fd8\u6709\u4e00\u4e2a\u5feb\u901f\u8f7b\u91cf\u7ea7\u7684\u63a9\u7801\u89e3\u7801\u5668\u3002\u8fd9\u79cd\u8bbe\u8ba1\u67b6\u6784\u80cc\u540e\u7684\u4e00\u4e2a\u91cd\u8981\u52a8\u673a\u662f\u5141\u8bb8\u5728\u8fb9\u7f18\u8bbe\u5907\u4e0a\uff08\u4f8b\u5982\u5728\u6d4f\u89c8\u5668\u4e2d\uff09\u8fdb\u884c\u5feb\u901f\u3001\u5b9e\u65f6\u7684\u5206\u5272\uff0c\u56e0\u4e3a\u56fe\u50cf\u5d4c\u5165\u53ea\u9700\u8981\u8ba1\u7b97\u4e00\u6b21\uff0c\u5e76\u4e14\u63a9\u7801\u89e3\u7801\u5668\u53ef\u4ee5\u5728CPU\u4e0a\u8fd0\u884c\u7ea650ms\u3002<\/p>\n<p>SAM\u7684\u6a21\u578b\u67b6\u6784\u5411\u6211\u4eec\u5c55\u793a\u4e86\u6a21\u578b\u63a5\u53d7\u54ea\u4e9b\u8f93\u5165\u4ee5\u53ca\u9700\u8981\u8bad\u7ec3\u6a21\u578b\u7684\u54ea\u4e9b\u90e8\u5206\uff08\u56fe\u7247\u6765\u6e90\u4e8eSAM GitHub\uff1ahttps:\/\/github.com\/facebookresearch\/segment-anything\uff09\u3002<\/p>\n<p>\u7406\u8bba\u4e0a\uff0c\u56fe\u50cf\u7f16\u7801\u5668\u5df2\u7ecf\u5b66\u4f1a\u4e86\u5d4c\u5165\u56fe\u50cf\u7684\u6700\u4f73\u65b9\u5f0f\uff0c\u5305\u62ec\u8bc6\u522b\u5f62\u72b6\u3001\u8fb9\u7f18\u548c\u5176\u4ed6\u4e00\u822c\u89c6\u89c9\u7279\u5f81\u7b49\u3002\u7c7b\u4f3c\u5730\uff0c\u5728\u7406\u8bba\u4e0a\uff0c\u63d0\u793a\u7f16\u7801\u5668\u5df2\u7ecf\u80fd\u591f\u4ee5\u6700\u4f18\u65b9\u5f0f\u5bf9\u63d0\u793a\u8fdb\u884c\u7f16\u7801\u3002\u63a9\u7801\u89e3\u7801\u5668\u662f\u6a21\u578b\u67b6\u6784\u7684\u4e00\u90e8\u5206\uff0c\u5b83\u91c7\u7528\u8fd9\u4e9b\u56fe\u50cf\u548c\u63d0\u793a\u5d4c\u5165\uff0c\u5e76\u901a\u8fc7\u5bf9\u56fe\u50cf\u548c\u63d0\u793a\u5d4c\u5165\u5f0f\u8fdb\u884c\u64cd\u4f5c\u6765\u5b9e\u9645\u521b\u5efa\u63a9\u7801\u3002<\/p>\n<p>\u56e0\u6b64\uff0c\u4e00\u79cd\u65b9\u6cd5\u662f\u5728\u8bad\u7ec3\u671f\u95f4\u51bb\u7ed3\u4e0e\u56fe\u50cf\u548c\u63d0\u793a\u7f16\u7801\u5668\u76f8\u5173\u8054\u7684\u6a21\u578b\u53c2\u6570\uff0c\u5e76\u4e14\u4ec5\u66f4\u65b0\u63a9\u7801\u89e3\u7801\u5668\u6743\u91cd\u3002\u8fd9\u79cd\u65b9\u6cd5\u7684\u4f18\u70b9\u662f\u5141\u8bb8\u6709\u76d1\u7763\u548c\u65e0\u76d1\u7763\u7684\u4e0b\u6e38\u4efb\u52a1\uff0c\u56e0\u4e3a\u63a7\u5236\u70b9\u548c\u8fb9\u754c\u6846\u63d0\u793a\u90fd\u662f\u81ea\u52a8\u7684\uff0c\u5e76\u4e14\u53ef\u4f9b\u4eba\u5de5\u4f7f\u7528\u3002<\/p>\n<p>\u56fe\u4e2d\u663e\u793a\u4e86AutoSAM\u4f53\u7cfb\u67b6\u6784\u4e2d\u4f7f\u7528\u7684\u51bb\u7ed3SAM\u56fe\u50cf\u7f16\u7801\u5668\u548c\u63a9\u7801\u89e3\u7801\u5668\uff0c\u4ee5\u53ca\u8fc7\u8f7d\u63d0\u793a\u7f16\u7801\u5668\uff08\u6765\u6e90\u4e8eAutoSAM\u8bba\u6587\uff1ahttps:\/\/arxiv.org\/pdf\/2306.06370\uff09\u3002<\/p>\n<p>\u53e6\u4e00\u79cd\u65b9\u6cd5\u662f\u4f7f\u63d0\u793a\u7f16\u7801\u5668\u8fc7\u8f7d\uff0c\u51bb\u7ed3\u56fe\u50cf\u7f16\u7801\u5668\u548c\u63a9\u7801\u89e3\u7801\u5668\uff0c\u5e76\u4e14\u53ea\u662f\u7b80\u5355\u5730\u4e0d\u4f7f\u7528\u539f\u59cbSAM\u63a9\u7801\u7f16\u7801\u5668\u3002\u4f8b\u5982\uff0cAutoSAM\u4f53\u7cfb\u67b6\u6784\u4f7f\u7528\u57fa\u4e8eHarmonic Dense Net\u7684\u7f51\u7edc\u6765\u57fa\u4e8e\u56fe\u50cf\u672c\u8eab\u751f\u6210\u63d0\u793a\u5d4c\u5165\u3002\u5728\u672c\u6559\u7a0b\u4e2d\uff0c\u6211\u4eec\u5c06\u4ecb\u7ecd\u7b2c\u4e00\u79cd\u65b9\u6cd5\uff0c\u5373\u51bb\u7ed3\u56fe\u50cf\u548c\u63d0\u793a\u7f16\u7801\u5668\uff0c\u53ea\u8bad\u7ec3\u63a9\u7801\u89e3\u7801\u5668\uff0c\u4f46\u8fd9\u79cd\u66ff\u4ee3\u65b9\u6cd5\u7684\u4ee3\u7801\u53ef\u4ee5\u5728AutoSAM GitHub\uff08https:\/\/github.com\/talshaharabany\/AutoSAM\/blob\/main\/inference.py\uff09\u548c\u8bba\u6587\uff08https:\/\/arxiv.org\/pdf\/2306.06370\uff09\u4e2d\u627e\u5230\u3002<\/p>\n<h3>\u914d\u7f6e\u63d0\u793a<\/h3>\n<p>\u63a5\u4e0b\u6765\u7684\u4e00\u6b65\u662f\u786e\u5b9a\u6a21\u578b\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u4f1a\u6536\u5230\u4ec0\u4e48\u7c7b\u578b\u7684\u63d0\u793a\uff0c\u4ee5\u4fbf\u6211\u4eec\u53ef\u4ee5\u5728\u8bad\u7ec3\u65f6\u63d0\u4f9b\u8fd9\u79cd\u7c7b\u578b\u7684\u63d0\u793a\u3002\u5c31\u6211\u4e2a\u4eba\u800c\u8a00\uff0c\u8003\u8651\u5230\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7684\u4e0d\u53ef\u9884\u6d4b\/\u4e0d\u4e00\u81f4\u6027\uff0c\u6211\u4e0d\u5efa\u8bae\u5728\u4efb\u4f55\u4e25\u8083\u7684\u8ba1\u7b97\u673a\u89c6\u89c9\u9879\u76ee\u67b6\u6784\u4e2d\u4f7f\u7528\u6587\u672c\u63d0\u793a\u3002\u5269\u4e0b\u7684\u89e3\u51b3\u65b9\u6848\u5c31\u9700\u8981\u4f9d\u8d56\u63a7\u5236\u70b9\u548c\u8fb9\u754c\u6846\u6280\u672f\u4e86\uff1b\u4f46\u662f\uff0c\u6700\u7ec8\u7684\u9009\u62e9\u8fd8\u8981\u53d6\u51b3\u4e8e\u7279\u5b9a\u6570\u636e\u96c6\u7684\u7279\u5b9a\u6027\u8d28\uff0c\u5c3d\u7ba1\u6709\u5173\u6587\u732e\u4e2d\u5df2\u7ecf\u6307\u51fa\u8fb9\u754c\u6846\u65b9\u6848\u7684\u8868\u73b0\u76f8\u5f53\u4e00\u81f4\u5730\u4f18\u4e8e\u63a7\u5236\u70b9\u65b9\u6848\u3002<\/p>\n<p>\u9020\u6210\u8fd9\u79cd\u60c5\u51b5\u7684\u539f\u56e0\u5c1a\u4e0d\u5b8c\u5168\u6e05\u695a\uff0c\u4f46\u53ef\u80fd\u662f\u4ee5\u4e0b\u4efb\u4f55\u56e0\u7d20\u4e4b\u4e00\uff0c\u6216\u8005\u662f\u8fd9\u4e9b\u56e0\u7d20\u7684\u7ec4\u5408\uff1a<\/p>\n<ul data-id=\"u738a58b-M4dVQF9C\">\n<li data-id=\"ld70c578-X5ViQmAL\">\u5728\u63a8\u7406\u65f6\uff08\u5f53\u771f\u5b9e\u503c\u63a9\u7801\u672a\u77e5\u65f6\uff09\uff0c\u597d\u7684\u63a7\u5236\u70b9\u6bd4\u8fb9\u754c\u6846\u66f4\u96be\u9009\u62e9\u3002<\/li>\n<li data-id=\"ld70c578-412eSaXV\">\u53ef\u80fd\u7684\u70b9\u63d0\u793a\u7684\u7a7a\u95f4\u6bd4\u53ef\u80fd\u7684\u8fb9\u754c\u6846\u63d0\u793a\u7684\u7a7a\u95f4\u5927\u51e0\u4e2a\u6570\u91cf\u7ea7\uff0c\u56e0\u6b64\u5b83\u6ca1\u6709\u7ecf\u8fc7\u5f7b\u5e95\u7684\u8bad\u7ec3\u3002<\/li>\n<li data-id=\"ld70c578-PnU9S2Vc\">\u6700\u521d\u7684SAM\u6a21\u578b\u4f5c\u8005\u4e3b\u8981\u4e13\u6ce8\u4e8e\u6a21\u578b\u7684\u96f6\u6837\u672c\u548c\u5c11\u6837\u672c\uff08\u6839\u636e\u4eba\u5de5\u63d0\u793a\u4ea4\u4e92\u8ba1\u7b97\uff09\u529f\u80fd\uff0c\u56e0\u6b64\u9884\u8bad\u7ec3\u53ef\u80fd\u66f4\u591a\u5730\u5173\u6ce8\u8fb9\u754c\u6846\u3002<\/li>\n<\/ul>\n<p>\u65e0\u8bba\u5982\u4f55\uff0c\u6cb3\u6d41\u5206\u5272\u5b9e\u9645\u4e0a\u662f\u4e00\u79cd\u7f55\u89c1\u7684\u60c5\u51b5\uff1b\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\uff0c\u70b9\u63d0\u793a\u65b9\u6848\u5b9e\u9645\u4e0a\u4f18\u4e8e\u8fb9\u754c\u6846\uff08\u5c3d\u7ba1\u53ea\u662f\u8f7b\u5fae\u7684\uff0c\u5373\u4f7f\u662f\u5728\u975e\u5e38\u6709\u5229\u7684\u57df\u4e2d\uff09\u3002\u5047\u8bbe\u5728\u6cb3\u6d41\u7684\u4efb\u4f55\u56fe\u50cf\u4e2d\uff0c\u6c34\u4f53\u5c06\u4ece\u56fe\u50cf\u7684\u4e00\u7aef\u5ef6\u4f38\u5230\u53e6\u4e00\u7aef\uff0c\u4efb\u4f55\u5305\u542b\u7684\u8fb9\u754c\u6846\u51e0\u4e4e\u603b\u662f\u8986\u76d6\u56fe\u50cf\u7684\u5927\u90e8\u5206\u3002\u56e0\u6b64\uff0c\u6cb3\u6d41\u975e\u5e38\u4e0d\u540c\u90e8\u5206\u7684\u8fb9\u754c\u6846\u63d0\u793a\u770b\u8d77\u6765\u975e\u5e38\u76f8\u4f3c\u3002\u7406\u8bba\u4e0a\uff0c\u8fd9\u610f\u5473\u7740\u8fb9\u754c\u6846\u4e3a\u6a21\u578b\u63d0\u4f9b\u7684\u4fe1\u606f\u6bd4\u63a7\u5236\u70b9\u5c11\u5f97\u591a\uff1b\u56e0\u6b64\uff0c\u5bfc\u81f4\u6027\u80fd\u8f83\u5dee\u3002<\/p>\n<p>\u63a7\u5236\u70b9\u3001\u8fb9\u754c\u6846\u63d0\u793a\u548c\u53e0\u52a0\u5728\u4e24\u4e2a\u6837\u672c\u8bad\u7ec3\u56fe\u50cf\u4e0a\u7684\u771f\u5b9e\u5206\u5272<\/p>\n<p>\u8bf7\u6ce8\u610f\uff0c\u5728\u4e0a\u56fe\u4e2d\uff0c\u5c3d\u7ba1\u4e24\u6761\u6cb3\u6d41\u90e8\u5206\u7684\u771f\u5b9e\u5206\u5272\u63a9\u7801\u5b8c\u5168\u4e0d\u540c\uff0c\u4f46\u5b83\u4eec\u5404\u81ea\u7684\u8fb9\u754c\u6846\u51e0\u4e4e\u76f8\u540c\uff0c\u800c\u5b83\u4eec\u7684\u70b9\u63d0\u793a\uff08\u76f8\u5bf9\u800c\u8a00\uff09\u5dee\u5f02\u66f4\u5927\u3002<\/p>\n<p>\u53e6\u4e00\u4e2a\u9700\u8981\u8003\u8651\u7684\u91cd\u8981\u56e0\u7d20\u662f\u5728\u63a8\u7406\u65f6\u751f\u6210\u8f93\u5165\u63d0\u793a\u7684\u5bb9\u6613\u7a0b\u5ea6\u3002\u5982\u679c\u60a8\u5e0c\u671b\u5728\u5faa\u73af\u6267\u884c\u9636\u6bb5\u6709\u4eba\u5de5\u4ecb\u5165\uff0c\u90a3\u4e48\u8bf7\u6ce8\u610f\u8fb9\u754c\u6846\u548c\u63a7\u5236\u70b9\u5728\u63a8\u7406\u9636\u6bb5\u90fd\u662f\u76f8\u5f53\u7410\u788e\u7684\u3002\u7136\u800c\uff0c\u5982\u679c\u60a8\u6253\u7b97\u4f7f\u7528\u4e00\u4e2a\u5b8c\u5168\u81ea\u52a8\u5316\u7684\u67b6\u6784\u65b9\u6848\uff0c\u90a3\u4e48\u56de\u7b54\u8fd9\u4e9b\u95ee\u9898\u5c06\u53d8\u5f97\u66f4\u52a0\u590d\u6742\u3002<\/p>\n<p>\u65e0\u8bba\u662f\u4f7f\u7528\u63a7\u5236\u70b9\u8fd8\u662f\u8fb9\u754c\u6846\uff0c\u751f\u6210\u63d0\u793a\u901a\u5e38\u9996\u5148\u5305\u62ec\u4f30\u8ba1\u611f\u5174\u8da3\u5bf9\u8c61\u7684\u7c97\u7565\u63a9\u7801\u3002\u8fb9\u754c\u6846\u53ef\u4ee5\u53ea\u662f\u5305\u88f9\u7c97\u7565\u63a9\u7801\u7684\u6700\u5c0f\u6846\uff0c\u800c\u63a7\u5236\u70b9\u9700\u8981\u4ece\u7c97\u7565\u63a9\u7801\u4e2d\u91c7\u6837\u3002\u8fd9\u610f\u5473\u7740\uff0c\u5f53\u771f\u5b9e\u503c\u63a9\u7801\u672a\u77e5\u65f6\uff0c\u8fb9\u754c\u6846\u66f4\u5bb9\u6613\u83b7\u5f97\uff0c\u56e0\u4e3a\u611f\u5174\u8da3\u5bf9\u8c61\u7684\u4f30\u8ba1\u63a9\u7801\u53ea\u9700\u8981\u5927\u81f4\u5339\u914d\u771f\u5b9e\u5bf9\u8c61\u7684\u76f8\u540c\u5927\u5c0f\u548c\u4f4d\u7f6e\uff1b\u800c\u5bf9\u4e8e\u63a7\u5236\u70b9\uff0c\u4f30\u8ba1\u63a9\u7801\u5c06\u9700\u8981\u66f4\u7d27\u5bc6\u5730\u5339\u914d\u5bf9\u8c61\u7684\u8f6e\u5ed3\u3002<\/p>\n<p>\u5f53\u4f7f\u7528\u4f30\u8ba1\u7684\u63a9\u7801\u800c\u4e0d\u662f\u771f\u5b9e\u503c\u65f6\uff0c\u63a7\u5236\u70b9\u7684\u653e\u7f6e\u53ef\u80fd\u5305\u62ec\u9519\u8bef\u6807\u6ce8\u7684\u70b9\uff0c\u800c\u8fb9\u754c\u6846\u901a\u5e38\u4f4d\u4e8e\u6b63\u786e\u7684\u4f4d\u7f6e<\/p>\n<p>\u5bf9\u4e8e\u6cb3\u6d41\u5206\u5272\uff0c\u5982\u679c\u6211\u4eec\u53ef\u4ee5\u540c\u65f6\u4f7f\u7528RGB\u548cNIR\uff0c\u90a3\u4e48\u6211\u4eec\u53ef\u4ee5\u4f7f\u7528\u5149\u8c31\u6307\u6570\u9608\u503c\u65b9\u6cd5\u6765\u83b7\u5f97\u6211\u4eec\u7684\u7c97\u7565\u63a9\u6a21\u3002\u5982\u679c\u6211\u4eec\u53ea\u80fd\u4f7f\u7528RGB\u6a21\u5f0f\uff0c\u6211\u4eec\u53ef\u4ee5\u5c06\u56fe\u50cf\u8f6c\u6362\u4e3aHSV\u6a21\u5f0f\uff0c\u5e76\u5bf9\u7279\u5b9a\u8272\u8c03\u3001\u9971\u548c\u5ea6\u548c\u503c\u8303\u56f4\u5185\u7684\u6240\u6709\u50cf\u7d20\u8bbe\u7f6e\u9608\u503c\u3002\u7136\u540e\uff0c\u6211\u4eec\u53ef\u4ee5\u79fb\u9664\u4f4e\u4e8e\u7279\u5b9a\u5927\u5c0f\u9608\u503c\u7684\u8fde\u63a5\u5185\u5bb9\uff0c\u5e76\u4f7f\u7528skimage.morphology\u5b50\u6a21\u5757\u4e2d\u7684erosion\u51fd\u6570\u6765\u786e\u4fdd\u6211\u4eec\u7684\u63a9\u6a21\u4e2d\u53ea\u67091\u4e2a\u50cf\u7d20\u662f\u671d\u5411\u84dd\u8272\u5927\u6591\u70b9\u4e2d\u5fc3\u7684\u50cf\u7d20\u3002<\/p>\n<h3>\u6a21\u578b\u8bad\u7ec3<\/h3>\n<p>\u4e3a\u4e86\u8bad\u7ec3\u6211\u4eec\u7684\u6a21\u578b\uff0c\u6211\u4eec\u9700\u8981\u4e00\u4e2a\u5305\u542b\u6240\u6709\u8bad\u7ec3\u6570\u636e\u7684\u6570\u636e\u52a0\u8f7d\u5668\uff0c\u6211\u4eec\u53ef\u4ee5\u5728\u6bcf\u4e2a\u8bad\u7ec3\u65f6\u671f\u5bf9\u8fd9\u4e9b\u6570\u636e\u8fdb\u884c\u8fed\u4ee3\u3002\u5f53\u6211\u4eec\u4eceHuggingFace\u52a0\u8f7d\u6570\u636e\u96c6\u65f6\uff0c\u5b83\u91c7\u7528datasets.Dataset\u7c7b\u7684\u5f62\u5f0f\u3002\u5982\u679c\u6570\u636e\u96c6\u662f\u79c1\u6709\u7684\uff0c\u8bf7\u786e\u4fdd\u9996\u5148\u5b89\u88c5HuggingFace CLI\u5e76\u4f7f\u7528\u201c!huggingface-cli login\u201d\u65b9\u5f0f\u767b\u5f55\u3002<\/p>\n<p>\u590d\u5236<\/p>\n<p>from datasets import load_dataset, load_from_disk, Dataset hf_dataset_name = &#8220;stodoran\/elwha-segmentation-v1&#8243; training_data = load_dataset(hf_dataset_name, split=&#8221;train&#8221;) validation_data = load_dataset(hf_dataset_name, split=&#8221;validation&#8221;)<\/p>\n<ul>\n<li>1.<\/li>\n<li>2.<\/li>\n<li>3.<\/li>\n<li>4.<\/li>\n<\/ul>\n<p>\u7136\u540e\uff0c\u6211\u4eec\u9700\u8981\u7f16\u5199\u81ea\u5df1\u7684\u81ea\u5b9a\u4e49\u6570\u636e\u96c6\u7c7b\uff0c\u8be5\u7c7b\u4e0d\u4ec5\u8fd4\u56de\u4efb\u4f55\u7d22\u5f15\u7684\u56fe\u50cf\u548c\u6807\u7b7e\uff0c\u8fd8\u8fd4\u56de\u63d0\u793a\u8bcd\u4fe1\u606f\u3002\u4e0b\u9762\u662f\u4e00\u4e2a\u53ef\u4ee5\u540c\u65f6\u5904\u7406\u63a7\u5236\u70b9\u548c\u8fb9\u754c\u6846\u63d0\u793a\u7684\u5b9e\u73b0\u3002\u8981\u5b8c\u6210\u521d\u59cb\u5316\u5de5\u4f5c\uff0c\u9700\u8981\u4e00\u4e2aHuggingFace datasets.Dataset\u5b9e\u4f8b\u548cSAM\u6a21\u578b\u7684\u5904\u7406\u5668\u5b9e\u4f8b\u3002<\/p>\n<p>\u590d\u5236<\/p>\n<p>from torch.utils.data import Dataset class PromptType: CONTROL_POINTS = &#8220;pts&#8221; BOUNDING_BOX = &#8220;bbox&#8221; class SAMDataset(Dataset): def __init__( self, dataset, processor, prompt_type = PromptType.CONTROL_POINTS, num_positive = 3, num_negative = 0, erode = True, multi_mask = &#8220;mean&#8221;, perturbation = 10, image_size = (1024, 1024), mask_size = (256, 256), ): #\u5c06\u6240\u6709\u503c\u8d4b\u7ed9self &#8230; def __len__(self): return len(self.dataset) def __getitem__(self, idx): datapoint = self.dataset[idx] input_image = cv2.resize(np.array(datapoint[&#8220;image&#8221;]), self.image_size) ground_truth_mask = cv2.resize(np.array(datapoint[&#8220;label&#8221;]), self.mask_size) if self.prompt_type == PromptType.CONTROL_POINTS: inputs = self._getitem_ctrlpts(input_image, ground_truth_mask) elif self.prompt_type == PromptType.BOUNDING_BOX: inputs = self._getitem_bbox(input_image, ground_truth_mask) inputs[&#8220;ground_truth_mask&#8221;] = ground_truth_mask return inputs<\/p>\n<ul>\n<li>1.<\/li>\n<li>2.<\/li>\n<li>3.<\/li>\n<li>4.<\/li>\n<li>5.<\/li>\n<li>6.<\/li>\n<li>7.<\/li>\n<li>8.<\/li>\n<li>9.<\/li>\n<li>10.<\/li>\n<li>11.<\/li>\n<li>12.<\/li>\n<li>13.<\/li>\n<li>14.<\/li>\n<li>15.<\/li>\n<li>16.<\/li>\n<li>17.<\/li>\n<li>18.<\/li>\n<li>19.<\/li>\n<li>20.<\/li>\n<li>21.<\/li>\n<li>22.<\/li>\n<li>23.<\/li>\n<li>24.<\/li>\n<li>25.<\/li>\n<li>26.<\/li>\n<li>27.<\/li>\n<li>28.<\/li>\n<li>29.<\/li>\n<li>30.<\/li>\n<li>31.<\/li>\n<li>32.<\/li>\n<li>33.<\/li>\n<li>34.<\/li>\n<li>35.<\/li>\n<li>36.<\/li>\n<\/ul>\n<p>\u6211\u4eec\u8fd8\u5fc5\u987b\u5b9a\u4e49SAMDataset_getitem_ctrlpts\u548cSAMDataset_getitem_box\u51fd\u6570\uff0c\u5c3d\u7ba1\u5982\u679c\u60a8\u53ea\u8ba1\u5212\u4f7f\u7528\u4e00\u79cd\u63d0\u793a\u7c7b\u578b\uff0c\u90a3\u4e48\u60a8\u53ef\u4ee5\u91cd\u6784\u4ee3\u7801\u4ee5\u76f4\u63a5\u5904\u7406SAMDataset.__getitem__\u4e2d\u7684\u8be5\u7c7b\u578b\uff0c\u5e76\u5220\u9664\u5e2e\u52a9\u7c7b\u5de5\u5177\u51fd\u6570\u3002<\/p>\n<p>\u590d\u5236<\/p>\n<p>class SAMDataset(Dataset): &#8230; def _getitem_ctrlpts(self, input_image, ground_truth_mask): # \u83b7\u53d6\u63a7\u5236\u70b9\u63d0\u793a\u3002\u8bf7\u53c2\u9605GitHub\u83b7\u53d6\u8be5\u51fd\u6570\u7684\u6e90\u4ee3\u7801\uff0c\u6216\u5c06\u5176\u66ff\u6362\u4e3a\u60a8\u81ea\u5df1\u7684\u70b9\u9009\u62e9\u7b97\u6cd5\u3002 input_points, input_labels = generate_input_points( num_positive=self.num_positive, num_negative=self.num_negative, mask=ground_truth_mask, dynamic_distance=True, erode=self.erode, ) input_points = input_points.astype(float).tolist() input_labels = input_labels.tolist() input_labels = [[x] for x in input_labels] # \u4e3a\u6a21\u578b\u51c6\u5907\u56fe\u50cf\u548c\u63d0\u793a\u3002 inputs = self.processor( input_image, input_points=input_points, input_labels=input_labels, return_tensors=&#8221;pt&#8221; ) #\u5220\u9664\u5904\u7406\u5668\u9ed8\u8ba4\u6dfb\u52a0\u7684\u6279\u6b21\u7ef4\u5ea6\u3002 inputs = {k: v.squeeze(0) for k, v in inputs.items()} inputs[&#8220;input_labels&#8221;] = inputs[&#8220;input_labels&#8221;].squeeze(1) return inputs def _getitem_bbox(self, input_image, ground_truth_mask): #\u83b7\u53d6\u8fb9\u754c\u6846\u63d0\u793a\u3002 bbox = get_input_bbox(ground_truth_mask, perturbation=self.perturbation) #\u4e3a\u6a21\u578b\u51c6\u5907\u56fe\u50cf\u548c\u63d0\u793a\u3002 inputs = self.processor(input_image, input_boxes=[[bbox]], return_tensors=&#8221;pt&#8221;) inputs = {k: v.squeeze(0) for k, v in inputs.items()} # \u5220\u9664\u5904\u7406\u5668\u9ed8\u8ba4\u6dfb\u52a0\u7684\u6279\u6b21\u7ef4\u5ea6\u3002 return inputs<\/p>\n<ul>\n<li>1.<\/li>\n<li>2.<\/li>\n<li>3.<\/li>\n<li>4.<\/li>\n<li>5.<\/li>\n<li>6.<\/li>\n<li>7.<\/li>\n<li>8.<\/li>\n<li>9.<\/li>\n<li>10.<\/li>\n<li>11.<\/li>\n<li>12.<\/li>\n<li>13.<\/li>\n<li>14.<\/li>\n<li>15.<\/li>\n<li>16.<\/li>\n<li>17.<\/li>\n<li>18.<\/li>\n<li>19.<\/li>\n<li>20.<\/li>\n<li>21.<\/li>\n<li>22.<\/li>\n<li>23.<\/li>\n<li>24.<\/li>\n<li>25.<\/li>\n<li>26.<\/li>\n<li>27.<\/li>\n<li>28.<\/li>\n<li>29.<\/li>\n<li>30.<\/li>\n<li>31.<\/li>\n<li>32.<\/li>\n<li>33.<\/li>\n<li>34.<\/li>\n<li>35.<\/li>\n<li>36.<\/li>\n<li>37.<\/li>\n<li>38.<\/li>\n<\/ul>\n<p>\u5c06\u6240\u6709\u8fd9\u4e9b\u529f\u80fd\u7ec4\u5408\u5230\u4e00\u8d77\uff0c\u6211\u4eec\u53ef\u4ee5\u521b\u5efa\u4e00\u4e2a\u51fd\u6570\uff0c\u8be5\u51fd\u6570\u5728\u7ed9\u5b9aHuggingFace\u6570\u636e\u96c6\u7684\u4efb\u4e00\u90e8\u5206\u7684\u60c5\u51b5\u4e0b\u521b\u5efa\u5e76\u8fd4\u56dePyTorch\u6570\u636e\u52a0\u8f7d\u5668\u3002\u7f16\u5199\u8fd4\u56de\u6570\u636e\u52a0\u8f7d\u5668\u7684\u51fd\u6570\uff0c\u800c\u4e0d\u4ec5\u4ec5\u662f\u7528\u76f8\u540c\u7684\u4ee3\u7801\u6267\u884c\u5355\u5143\uff0c\u8fd9\u4e0d\u4ec5\u662f\u7f16\u5199\u7075\u6d3b\u548c\u53ef\u7ef4\u62a4\u4ee3\u7801\u7684\u597d\u65b9\u6cd5\uff0c\u800c\u4e14\u5982\u679c\u60a8\u8ba1\u5212\u4f7f\u7528HuggingFace Accelerate\uff08https:\/\/huggingface.co\/docs\/accelerate\/index\uff09\u6765\u8fd0\u884c\u5206\u5e03\u5f0f\u8bad\u7ec3\u7684\u8bdd\uff0c\u8fd9\u4e5f\u662f\u5fc5\u8981\u7684\u3002<\/p>\n<p>\u590d\u5236<\/p>\n<p>from transformers import SamProcessor from torch.utils.data import DataLoader def get_dataloader( hf_dataset, model_size = &#8220;base&#8221;, # One of &#8220;base&#8221;, &#8220;large&#8221;, or &#8220;huge&#8221; batch_size = 8, prompt_type = PromptType.CONTROL_POINTS, num_positive = 3, num_negative = 0, erode = True, multi_mask = &#8220;mean&#8221;, perturbation = 10, image_size = (256, 256), mask_size = (256, 256), ): processor = SamProcessor.from_pretrained(f&#8221;facebook\/sam-vit-{model_size}&#8221;) sam_dataset = SAMDataset( dataset=hf_dataset, processor=processor, prompt_type=prompt_type, num_positive=num_positive, num_negative=num_negative, erode=erode, multi_mask=multi_mask, perturbation=perturbation, image_size=image_size, mask_size=mask_size, ) dataloader = DataLoader(sam_dataset, batch_size=batch_size, shuffle=True) return dataloader<\/p>\n<ul>\n<li>1.<\/li>\n<li>2.<\/li>\n<li>3.<\/li>\n<li>4.<\/li>\n<li>5.<\/li>\n<li>6.<\/li>\n<li>7.<\/li>\n<li>8.<\/li>\n<li>9.<\/li>\n<li>10.<\/li>\n<li>11.<\/li>\n<li>12.<\/li>\n<li>13.<\/li>\n<li>14.<\/li>\n<li>15.<\/li>\n<li>16.<\/li>\n<li>17.<\/li>\n<li>18.<\/li>\n<li>19.<\/li>\n<li>20.<\/li>\n<li>21.<\/li>\n<li>22.<\/li>\n<li>23.<\/li>\n<li>24.<\/li>\n<li>25.<\/li>\n<li>26.<\/li>\n<li>27.<\/li>\n<li>28.<\/li>\n<li>29.<\/li>\n<li>30.<\/li>\n<li>31.<\/li>\n<li>32.<\/li>\n<li>33.<\/li>\n<\/ul>\n<p>\u5728\u6b64\u4e4b\u540e\uff0c\u8bad\u7ec3\u53ea\u9700\u52a0\u8f7d\u6a21\u578b\u3001\u51bb\u7ed3\u56fe\u50cf\u548c\u63d0\u793a\u7f16\u7801\u5668\uff0c\u5e76\u8fdb\u884c\u6240\u9700\u6b21\u6570\u7684\u8fed\u4ee3\u8bad\u7ec3\u3002<\/p>\n<p>\u590d\u5236<\/p>\n<p>model = SamModel.from_pretrained(f&#8221;facebook\/sam-vit-{model_size}&#8221;) optimizer = AdamW(model.mask_decoder.parameters(), lr=learning_rate, weight_decay=weight_decay) # Train only the decoder. for name, param in model.named_parameters(): if name.startswith(&#8220;vision_encoder&#8221;) or name.startswith(&#8220;prompt_encoder&#8221;): param.requires_grad_(False)<\/p>\n<ul>\n<li>1.<\/li>\n<li>2.<\/li>\n<li>3.<\/li>\n<li>4.<\/li>\n<li>5.<\/li>\n<li>6.<\/li>\n<li>7.<\/li>\n<\/ul>\n<p>\u4ee5\u4e0b\u5217\u51fa\u7684\u662f\u8bad\u7ec3\u8fc7\u7a0b\u5faa\u73af\u90e8\u5206\u7684\u57fa\u672c\u6846\u67b6\u4ee3\u7801\u3002\u8bf7\u6ce8\u610f\uff0c\u4e3a\u4e86\u7b80\u6d01\u8d77\u89c1\uff0cforward_pass\u3001calculate loss\u3001evaluate_mode\u548csave_model_checkpoint\u51fd\u6570\u88ab\u7701\u7565\u4e86\uff0c\u4f46GitHub\u4e0a\u63d0\u4f9b\u4e86\u5b9e\u73b0\u3002\u6b63\u5411\u4f20\u9012\u7801\u6839\u636e\u63d0\u793a\u7c7b\u578b\u7565\u6709\u4e0d\u540c\uff0c\u635f\u5931\u8ba1\u7b97\u4e5f\u9700\u8981\u57fa\u4e8e\u63d0\u793a\u7c7b\u578b\u7684\u7279\u6b8a\u60c5\u51b5\uff1b\u5f53\u4f7f\u7528\u70b9\u63d0\u793a\u65f6\uff0cSAM\u6a21\u578b\u4e3a\u6bcf\u4e2a\u5355\u4e2a\u8f93\u5165\u70b9\u8fd4\u56de\u4e00\u4e2a\u9884\u6d4b\u63a9\u7801\uff0c\u56e0\u6b64\u4e3a\u4e86\u83b7\u5f97\u53ef\u4ee5\u4e0e\u771f\u5b9e\u6570\u636e\u8fdb\u884c\u6bd4\u8f83\u7684\u5355\u4e2a\u63a9\u7801\uff0c\u9700\u8981\u5bf9\u9884\u6d4b\u63a9\u7801\u8fdb\u884c\u5e73\u5747\uff0c\u6216\u8005\u9700\u8981\u9009\u62e9\u6700\u4f73\u9884\u6d4b\u63a9\u7801\uff08\u57fa\u4e8eSAM\u7684\u9884\u6d4bIoU\u5206\u6570\u6765\u8bc6\u522b\uff09\u3002<\/p>\n<p>\u590d\u5236<\/p>\n<p>train_losses = [] validation_losses = [] epoch_loop = tqdm(total=num_epochs, position=epoch, leave=False) batch_loop = tqdm(total=len(train_dataloader), position=0, leave=True) while epoch &lt; num_epochs: epoch_losses = [] batch_loop.n = 0 #\u5faa\u73af\u91cd\u7f6e for idx, batch in enumerate(train_dataloader): # \u6b63\u5411\u4f20\u9012 batch = {k: v.to(accelerator.device) for k, v in batch.items()} outputs = forward_pass(model, batch, prompt_type) #\u8ba1\u7b97\u635f\u5931\u503c ground_truth_masks = batch[&#8220;ground_truth_mask&#8221;].float() train_loss = calculate_loss(outputs, ground_truth_masks, prompt_type, loss_fn, multi_mask=&#8221;best&#8221;) epoch_losses.append(train_loss) # \u53cd\u5411\u4f20\u9012\u4e0e\u4f18\u5316\u73af\u8282 optimizer.zero_grad() accelerator.backward(train_loss) optimizer.step() lr_scheduler.step() batch_loop.set_description(f&#8221;Train Loss: {train_loss.item():.4f}&#8221;) batch_loop.update(1) validation_loss = evaluate_model(model, validation_dataloader, accelerator.device, loss_fn) train_losses.append(torch.mean(torch.Tensor(epoch_losses))) validation_losses.append(validation_loss) if validation_loss &lt; best_loss: save_model_checkpoint( accelerator, best_checkpoint_path, model, optimizer, lr_scheduler, epoch, train_history, validation_loss, train_losses, validation_losses, loss_config, model_descriptor=model_descriptor, ) best_loss = validation_loss epoch_loop.set_description(f&#8221;Best Loss: {best_loss:.4f}&#8221;) epoch_loop.update(1) epoch += 1<\/p>\n<ul>\n<li>1.<\/li>\n<li>2.<\/li>\n<li>3.<\/li>\n<li>4.<\/li>\n<li>5.<\/li>\n<li>6.<\/li>\n<li>7.<\/li>\n<li>8.<\/li>\n<li>9.<\/li>\n<li>10.<\/li>\n<li>11.<\/li>\n<li>12.<\/li>\n<li>13.<\/li>\n<li>14.<\/li>\n<li>15.<\/li>\n<li>16.<\/li>\n<li>17.<\/li>\n<li>18.<\/li>\n<li>19.<\/li>\n<li>20.<\/li>\n<li>21.<\/li>\n<li>22.<\/li>\n<li>23.<\/li>\n<li>24.<\/li>\n<li>25.<\/li>\n<li>26.<\/li>\n<li>27.<\/li>\n<li>28.<\/li>\n<li>29.<\/li>\n<li>30.<\/li>\n<li>31.<\/li>\n<li>32.<\/li>\n<li>33.<\/li>\n<li>34.<\/li>\n<li>35.<\/li>\n<li>36.<\/li>\n<li>37.<\/li>\n<li>38.<\/li>\n<li>39.<\/li>\n<li>40.<\/li>\n<li>41.<\/li>\n<li>42.<\/li>\n<li>43.<\/li>\n<li>44.<\/li>\n<li>45.<\/li>\n<li>46.<\/li>\n<li>47.<\/li>\n<li>48.<\/li>\n<li>49.<\/li>\n<li>50.<\/li>\n<li>51.<\/li>\n<li>52.<\/li>\n<\/ul>\n<h3>\u5fae\u8c03\u7ed3\u679c\u5206\u6790<\/h3>\n<p>\u5bf9\u4e8e\u827e\u5c14\u74e6\u6cb3\u9879\u76ee\uff0c\u6700\u4f73\u8bbe\u7f6e\u662f\u5728\u4e0d\u523012\u5c0f\u65f6\u7684\u65f6\u95f4\u5185\u4f7f\u7528GCP\u865a\u62df\u673a\u5b9e\u4f8b\uff0c\u4f7f\u7528\u8d85\u8fc71k\u4e2a\u5206\u5272\u63a9\u7801\u7684\u6570\u636e\u96c6\u8bad\u7ec3\u6210\u529f\u201csam-vit-base\u201d\u6a21\u578b\u3002<\/p>\n<p>\u4e0e\u57fa\u51c6\u578bSAM\u76f8\u6bd4\uff0c\u5fae\u8c03\u663e\u8457\u63d0\u9ad8\u4e86\u6027\u80fd\uff0c\u4e2d\u503c\u63a9\u7801\u4ece\u4e0d\u53ef\u7528\u53d8\u4e3a\u9ad8\u5ea6\u51c6\u786e\u3002<\/p>\n<p>\u76f8\u5bf9\u4e8e\u57fa\u4e8e\u9ed8\u8ba4\u63d0\u793a\u8bcd\u7684\u57fa\u51c6\u578bSAM\u6a21\u578b\uff0c\u5fae\u8c03\u540e\u7684SAM\u6a21\u578b\u6781\u5927\u5730\u63d0\u9ad8\u4e86\u5206\u5272\u6027\u80fd<\/p>\n<p>\u9700\u8981\u6ce8\u610f\u7684\u4e00\u4e2a\u91cd\u8981\u4e8b\u5b9e\u662f\uff0c1k\u6cb3\u6d41\u56fe\u50cf\u7684\u8bad\u7ec3\u6570\u636e\u96c6\u662f\u4e0d\u5b8c\u7f8e\u7684\uff0c\u5206\u5272\u6807\u7b7e\u5728\u6b63\u786e\u5206\u7c7b\u7684\u50cf\u7d20\u6570\u91cf\u4e0a\u53d8\u5316\u5f88\u5927\u3002\u56e0\u6b64\uff0c\u4e0a\u8ff0\u6307\u6807\u662f\u5728225\u5e45\u6cb3\u6d41\u56fe\u50cf\u7684\u50cf\u7d20\u5b8c\u7f8e\u6570\u636e\u96c6\u4e0a\u8ba1\u7b97\u51fa\u6765\u7684\u3002<\/p>\n<p>\u5b9e\u9a8c\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u89c2\u5bdf\u5230\u7684\u4e00\u4e2a\u6709\u8da3\u7684\u884c\u4e3a\u662f\uff0c\u6a21\u578b\u5b66\u4f1a\u4e86\u4ece\u4e0d\u5b8c\u7f8e\u7684\u8bad\u7ec3\u6570\u636e\u4e2d\u8fdb\u884c\u5f52\u7eb3\u3002\u5f53\u5728\u8bad\u7ec3\u6837\u672c\u5305\u542b\u660e\u663e\u9519\u8bef\u5206\u7c7b\u7684\u6570\u636e\u70b9\u4e0a\u8fdb\u884c\u8bc4\u4f30\u65f6\uff0c\u6211\u4eec\u53ef\u4ee5\u89c2\u5bdf\u5230\u6a21\u578b\u9884\u6d4b\u907f\u514d\u4e86\u8bef\u5dee\u3002\u8bf7\u6ce8\u610f\uff0c\u663e\u793a\u8bad\u7ec3\u6837\u672c\u7684\u9876\u884c\u4e2d\u7684\u56fe\u50cf\u5305\u542b\u7684\u63a9\u7801\u4e0d\u4f1a\u4e00\u76f4\u586b\u5145\u5230\u6cb3\u5cb8\uff0c\u800c\u663e\u793a\u6a21\u578b\u9884\u6d4b\u7684\u5e95\u884c\u5219\u66f4\u7d27\u5bc6\u5730\u5206\u5272\u6cb3\u6d41\u8fb9\u754c\u3002<\/p>\n<p>\u5373\u4f7f\u8bad\u7ec3\u6570\u636e\u4e0d\u5b8c\u7f8e\uff0c\u7ecf\u5fae\u8c03\u7684SAM\u6a21\u578b\u4e5f\u80fd\u5e26\u6765\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\u7684\u6cdb\u5316\u6548\u679c\u3002\u8bf7\u6ce8\u610f\uff0c\u4e0e\u8bad\u7ec3\u6570\u636e\uff08\u9876\u884c\uff09\u76f8\u6bd4\uff0c\u9884\u6d4b\uff08\u5e95\u884c\uff09\u7684\u9519\u8bef\u5206\u7c7b\u66f4\u5c11\uff0c\u5e76\u4e14\u6cb3\u6d41\u7684\u586b\u5145\u7a0b\u5ea6\u66f4\u9ad8\u3002<\/p>\n<h3>\u7ed3\u8bba<\/h3>\n<p>\u5982\u679c\u60a8\u5df2\u7ecf\u987a\u5229\u5b8c\u6210\u672c\u6587\u4e2d\u7684\u5b9e\u4f8b\u5185\u5bb9\uff0c\u90a3\u4e48\u795d\u8d3a\u60a8\uff01\u60a8\u5df2\u7ecf\u5b66\u4f1a\u4e86\u4e3a\u4efb\u4f55\u4e0b\u6e38\u613f\u666f\u4efb\u52a1\u5b8c\u5168\u5fae\u8c03Meta\u7684\u5206\u5272\u4e00\u5207\u6a21\u578bSAM\u6240\u9700\u7684\u4e00\u5207\uff01<\/p>\n<p>\u867d\u7136\u60a8\u7684\u5fae\u8c03\u5de5\u4f5c\u6d41\u7a0b\u65e0\u7591\u4e0e\u672c\u6559\u7a0b\u4e2d\u4ecb\u7ecd\u7684\u5b9e\u65bd\u65b9\u5f0f\u4e0d\u540c\uff0c\u4f46\u4ece\u9605\u8bfb\u672c\u6559\u7a0b\u4e2d\u83b7\u5f97\u7684\u77e5\u8bc6\u4e0d\u4ec5\u4f1a\u5f71\u54cd\u5230\u60a8\u7684\u7ec6\u5206\u9879\u76ee\uff0c\u8fd8\u4f1a\u5f71\u54cd\u5230\u672a\u6765\u7684\u6df1\u5ea6\u5b66\u4e60\u9879\u76ee\u53ca\u5176\u4ed6\u9879\u76ee\u3002<\/p>\n<p>\u6700\u540e\uff0c\u5e0c\u671b\u60a8\u7ee7\u7eed\u63a2\u7d22\u673a\u5668\u5b66\u4e60\u7684\u4e16\u754c\uff0c\u4fdd\u6301\u597d\u5947\u5fc3\uff0c\u5e76\u4e00\u5982\u65e2\u5f80\u5730\u5feb\u4e50\u7f16\u7a0b\uff01<\/p>\n<h3>\u9644\u5f55<\/h3>\n<p>\u672c\u6587\u5b9e\u4f8b\u4e2d\u4f7f\u7528\u7684\u6570\u636e\u96c6\u662fElwha V1\u6570\u636e\u96c6\uff08https:\/\/huggingface.co\/datasets\/stodoran\/elwha-segmentation-v1\uff09\uff0c\u8be5\u6570\u636e\u96c6\u7531\u534e\u76db\u987f\u5927\u5b66\u7684GeoSMART\u7814\u7a76\u5b9e\u9a8c\u5ba4\uff08https:\/\/geo-smart.github.io\/\uff09\u521b\u5efa\uff0c\u7528\u4e8e\u5c06\u5fae\u8c03\u7684\u5927\u578b\u89c6\u89c9\u53d8\u6362\u5668\u5e94\u7528\u4e8e\u5730\u7406\u7a7a\u95f4\u5206\u5272\u4efb\u52a1\u7684\u7814\u7a76\u9879\u76ee\u3002\u672c\u6587\u63cf\u8ff0\u7684\u5185\u5bb9\u4ee3\u8868\u4e86\u5373\u5c06\u53d1\u8868\u7684\u8bba\u6587\u7684\u7cbe\u7b80\u7248\u548c\u4e00\u4e2a\u66f4\u6613\u4e8e\u5b9e\u73b0\u7684\u7248\u672c\u3002\u5728\u9ad8\u6c34\u5e73\u4e0a\uff0cElwha V1\u6570\u636e\u96c6\u7531SAM\u68c0\u67e5\u70b9\u7684\u540e\u5904\u7406\u6a21\u578b\u9884\u6d4b\u7ec4\u6210\uff0c\u8be5\u68c0\u67e5\u70b9\u4f7f\u7528Buscombe\u7b49\u4eba\uff08https:\/\/zenodo.org\/records\/10155783\uff09\u53d1\u5e03\u5e76\u5728\u591a\u5b66\u79d1\u7814\u7a76\u6570\u636e\u77e5\u8bc6\u5e93\u548c\u6587\u732e\u8d44\u6e90\u7f51\u7ad9Zenodo\u4e0a\u53d1\u5e03\u7684\u6807\u6ce8\u6b63\u5c04\u5f71\u50cf\u7684\u5b50\u96c6\u8fdb\u884c\u4e86\u5fae\u8c03\u3002<\/p>\n<h3>\u8bd1\u8005\u4ecb\u7ecd<\/h3>\n<p>\u6731\u5148\u5fe0\uff0c51CTO\u793e\u533a\u7f16\u8f91\uff0c51CTO\u4e13\u5bb6\u535a\u5ba2\u3001\u8bb2\u5e08\uff0c\u6f4d\u574a\u4e00\u6240\u9ad8\u6821\u8ba1\u7b97\u673a\u6559\u5e08\uff0c\u81ea\u7531\u7f16\u7a0b\u754c\u8001\u5175\u4e00\u679a\u3002<\/p>\n<p>\u539f\u6587\u6807\u9898\uff1aLearn Transformer Fine-Tuning and Segment Anything\uff0c\u4f5c\u8005\uff1aStefan Todoran<\/p>\n<p>\u94fe\u63a5\uff1ahttps:\/\/towardsdatascience.com\/learn-transformer-fine-tuning-and-segment-anything-481c6c4ac802<\/p>\n<p>\u60f3\u4e86\u89e3\u66f4\u591aAIGC\u7684\u5185\u5bb9\uff0c\u8bf7\u8bbf\u95ee\uff1a<\/p>\n<p>51CTO AI.x\u793e\u533a<\/p>\n<p>https:\/\/www.51cto.com\/aigc\/<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_21162\" class=\"pvc_stats total_only  \" data-element-id=\"21162\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u672c\u6587\u9996\u5148\u4ecb\u7ecdMeta\u516c\u53f8\u5f00\u53d1\u7684\u5f00\u6e90\u56fe\u50cf\u5206\u5272\u6a21\u578bSAM\u7684\u67b6\u6784\uff0c\u7136\u540e\u901a\u8fc7\u4e00\u4e2a\u6cb3\u6d41\u50cf\u7d20\u5206\u5272\u9065\u611f\u4efb\u52a1\u7684\u5b9e\u6218\u6848\u4f8b\u5c55\u793aSA [&hellip;]<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_21162\" class=\"pvc_stats total_only  \" data-element-id=\"21162\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"author":56,"featured_media":21166,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[23,17,80],"tags":[113,1204,842],"class_list":["post-21162","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-23","category-17","category-80","tag-113","tag-1204","tag-842"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/21162","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=21162"}],"version-history":[{"count":2,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/21162\/revisions"}],"predecessor-version":[{"id":21167,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/21162\/revisions\/21167"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/21166"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=21162"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=21162"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=21162"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}