{"id":26478,"date":"2025-02-23T13:21:07","date_gmt":"2025-02-23T05:21:07","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=26478"},"modified":"2025-02-23T13:21:55","modified_gmt":"2025-02-23T05:21:55","slug":"yolov12","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/yolov12\/","title":{"rendered":"YOLOv12\uff1a\u5b9e\u65f6\u56fe\u50cf\u548c\u89c6\u9891\u76ee\u6807\u68c0\u6d4b\u7684\u5f00\u6e90\u5de5\u5177"},"content":{"rendered":"<p>YOLOv12 \u662f\u7531 GitHub \u7528\u6237 sunsmarterjie \u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u4e13\u6ce8\u4e8e\u5b9e\u65f6\u76ee\u6807\u68c0\u6d4b\u6280\u672f\u3002\u8be5\u9879\u76ee\u57fa\u4e8e YOLO\uff08You Only Look Once\uff09\u7cfb\u5217\u6846\u67b6\uff0c\u5f15\u5165\u6ce8\u610f\u529b\u673a\u5236\u4f18\u5316\u4f20\u7edf\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08CNN\uff09\u7684\u6027\u80fd\uff0c\u4e0d\u4ec5\u5728\u68c0\u6d4b\u7cbe\u5ea6\u4e0a\u6709\u6240\u63d0\u5347\uff0c\u8fd8\u4fdd\u6301\u4e86\u9ad8\u6548\u7684\u63a8\u7406\u901f\u5ea6\u3002YOLOv12 \u9002\u7528\u4e8e\u591a\u79cd\u573a\u666f\uff0c\u5982\u76d1\u63a7\u7cfb\u7edf\u3001\u81ea\u52a8\u9a7e\u9a76\u548c\u56fe\u50cf\u5206\u6790\u7b49\uff0c\u63d0\u4f9b Nano\u3001Small\u3001Medium\u3001Large\u3001Extra-Large \u4e94\u79cd\u6a21\u578b\u89c4\u6a21\uff0c\u6ee1\u8db3\u4e0d\u540c\u8ba1\u7b97\u80fd\u529b\u548c\u5e94\u7528\u9700\u6c42\u3002\u9879\u76ee\u91c7\u7528 GNU AGPL-3.0 \u8bb8\u53ef\u8bc1\uff0c\u7528\u6237\u53ef\u4ee5\u514d\u8d39\u4e0b\u8f7d\u4ee3\u7801\u5e76\u6839\u636e\u9700\u6c42\u8fdb\u884c\u5b9a\u5236\u5f00\u53d1\u3002\u5f00\u53d1\u8005\u56e2\u961f\u5305\u62ec\u6765\u81ea\u5e03\u6cd5\u7f57\u5927\u5b66\u548c\u4e2d\u79d1\u9662\u7684\u7814\u7a76\u4eba\u5458\uff0c\u6280\u672f\u6587\u6863\u548c\u5b89\u88c5\u6307\u5357\u8be6\u5c3d\uff0c\u4fbf\u4e8e\u7528\u6237\u5feb\u901f\u4e0a\u624b\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-26479\" title=\"YOLOv12\uff1a\u63d0\u4f9b\u5b9e\u65f6\u56fe\u50cf\u548c\u89c6\u9891\u76ee\u6807\u68c0\u6d4b\u7684\u5f00\u6e90\u5de5\u5177-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/ac2c5ff1a8eb2a0.jpg\" alt=\"YOLOv12\uff1a\u63d0\u4f9b\u5b9e\u65f6\u56fe\u50cf\u548c\u89c6\u9891\u76ee\u6807\u68c0\u6d4b\u7684\u5f00\u6e90\u5de5\u5177-1\" width=\"1000\" height=\"558\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/ac2c5ff1a8eb2a0.jpg 1000w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/ac2c5ff1a8eb2a0-768x429.jpg 768w\" sizes=\"auto, (max-width: 1000px) 100vw, 1000px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u9ad8\u6548\u5b9e\u65f6\u76ee\u6807\u68c0\u6d4b<\/strong>: \u5728 T4 GPU \u4e0a\uff0cYOLOv12-N \u53ef\u5b9e\u73b0 40.6% mAP\uff0c\u63a8\u7406\u5ef6\u8fdf\u4ec5 1.64ms\u3002<\/li>\n<li><strong>\u591a\u6a21\u578b\u9009\u62e9<\/strong>: \u63d0\u4f9b\u4e94\u79cd\u6a21\u578b\uff08Nano \u5230 Extra-Large\uff09\uff0c\u9002\u914d\u4ece\u4f4e\u529f\u8017\u8bbe\u5907\u5230\u9ad8\u6027\u80fd\u670d\u52a1\u5668\u7684\u591a\u79cd\u786c\u4ef6\u73af\u5883\u3002<\/li>\n<li><strong>\u6ce8\u610f\u529b\u673a\u5236\u4f18\u5316<\/strong>: \u5f15\u5165\u201c\u533a\u57df\u6ce8\u610f\u529b\u201d\uff08Area Attention\uff09\u548c R-ELAN \u6a21\u5757\uff0c\u63d0\u5347\u68c0\u6d4b\u7cbe\u5ea6\u5e76\u51cf\u5c11\u8ba1\u7b97\u590d\u6742\u5ea6\u3002<\/li>\n<li><strong>\u6a21\u578b\u5bfc\u51fa<\/strong>: \u652f\u6301\u5c06\u8bad\u7ec3\u6a21\u578b\u5bfc\u51fa\u4e3a ONNX \u6216 TensorRT \u683c\u5f0f\uff0c\u65b9\u4fbf\u90e8\u7f72\u5230\u751f\u4ea7\u73af\u5883\u3002<\/li>\n<li><strong>\u81ea\u5b9a\u4e49\u6570\u636e\u96c6\u8bad\u7ec3<\/strong>: \u7528\u6237\u53ef\u4f7f\u7528\u81ea\u5df1\u7684\u6570\u636e\u96c6\u8bad\u7ec3\u6a21\u578b\uff0c\u9002\u7528\u4e8e\u7279\u5b9a\u76ee\u6807\u68c0\u6d4b\u4efb\u52a1\u3002<\/li>\n<li><strong>\u53ef\u89c6\u5316\u652f\u6301<\/strong>: \u96c6\u6210\u76d1\u7763\u5de5\u5177\uff08supervision\uff09\uff0c\u65b9\u4fbf\u5c55\u793a\u68c0\u6d4b\u7ed3\u679c\u548c\u6027\u80fd\u8bc4\u4f30\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<p>YOLOv12 \u76ee\u524d\u6ca1\u6709\u72ec\u7acb\u7684 PyPI \u5305\uff0c\u9700\u8981\u4ece GitHub \u6e90\u7801\u5b89\u88c5\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u5b89\u88c5\u6b65\u9aa4\uff0c\u9002\u7528\u4e8e Linux \u7cfb\u7edf\uff08Windows \u6216 Mac \u7528\u6237\u9700\u8c03\u6574\u73af\u5883\u914d\u7f6e\uff09\uff1a<\/p>\n<ol>\n<li><strong>\u51c6\u5907\u73af\u5883<\/strong>\n<ul>\n<li>\u786e\u4fdd\u7cfb\u7edf\u5b89\u88c5 Python 3.11 \u6216\u66f4\u9ad8\u7248\u672c\u3002<\/li>\n<li>\u5b89\u88c5 Git\uff1a<code>sudo apt install git<\/code>\uff08Ubuntu \u793a\u4f8b\uff09\u3002<\/li>\n<li>\u53ef\u9009\uff1a\u5b89\u88c5 NVIDIA GPU \u9a71\u52a8\u548c CUDA\uff08\u63a8\u8350 11.8 \u6216\u66f4\u9ad8\u7248\u672c\uff09\u4ee5\u52a0\u901f\u8bad\u7ec3\u548c\u63a8\u7406\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u514b\u9686\u4ed3\u5e93<\/strong><br \/>\n\u5728\u7ec8\u7aef\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\uff0c\u5c06 YOLOv12 \u4ed3\u5e93\u4e0b\u8f7d\u5230\u672c\u5730\uff1a<\/p>\n<pre><code>git clone https:\/\/github.com\/sunsmarterjie\/yolov12.git\r\ncd yolov12<\/code><\/pre>\n<\/li>\n<\/ol>\n<ol start=\"3\">\n<li><strong>\u521b\u5efa\u865a\u62df\u73af\u5883<\/strong><br \/>\n\u4f7f\u7528 Conda \u6216 venv \u521b\u5efa\u72ec\u7acb\u7684 Python \u73af\u5883\uff0c\u907f\u514d\u4f9d\u8d56\u51b2\u7a81\uff1a<\/p>\n<pre><code>conda create -n yolov12 python=3.11\r\nconda activate yolov12\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5b89\u88c5\u4f9d\u8d56<\/strong><br \/>\n\u5b89\u88c5\u9879\u76ee\u6240\u9700\u7684\u4f9d\u8d56\u5305\uff0c\u5305\u62ec PyTorch\u3001flash-attn \u548c supervision \u7b49\uff1a<\/p>\n<pre><code>wget https:\/\/github.com\/Dao-AILab\/flash-attention\/releases\/download\/v2.7.3\/flash_attn-2.7.3+cu11torch2.2cxx11abiFALSE-cp311-cp311-linux_x86_64.whl\r\npip install flash_attn-2.7.3+cu11torch2.2cxx11abiFALSE-cp311-cp311-linux_x86_64.whl\r\npip install -r requirements.txt\r\npip install -e .\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u9a8c\u8bc1\u5b89\u88c5<\/strong><br \/>\n\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u68c0\u67e5\u73af\u5883\u662f\u5426\u6b63\u786e\u914d\u7f6e\uff1a<\/p>\n<pre><code>python -c \"from ultralytics import YOLO; print('YOLOv12 installed successfully')\"\r\n<\/code><\/pre>\n<\/li>\n<\/ol>\n<h3>\u4f7f\u7528\u65b9\u6cd5<\/h3>\n<h4>\u8bad\u7ec3\u81ea\u5b9a\u4e49\u6a21\u578b<\/h4>\n<p>YOLOv12 \u652f\u6301\u7528\u6237\u4f7f\u7528\u81ea\u5df1\u7684\u6570\u636e\u96c6\u8fdb\u884c\u8bad\u7ec3\uff0c\u9002\u5408\u7279\u5b9a\u573a\u666f\u7684\u76ee\u6807\u68c0\u6d4b\u4efb\u52a1\u3002\u64cd\u4f5c\u6b65\u9aa4\u5982\u4e0b\uff1a<\/p>\n<ol>\n<li><strong>\u51c6\u5907\u6570\u636e\u96c6<\/strong>\n<ul>\n<li>\u6570\u636e\u9700\u7b26\u5408 YOLO \u683c\u5f0f\uff08\u5305\u542b images \u548c labels \u6587\u4ef6\u5939\uff0clabels \u4e3a .txt \u6587\u4ef6\uff0c\u6807\u6ce8\u76ee\u6807\u7c7b\u522b\u548c\u8fb9\u754c\u6846\u5750\u6807\uff09\u3002<\/li>\n<li>\u521b\u5efa\u00a0<code>data.yaml<\/code>\u00a0\u6587\u4ef6\uff0c\u6307\u5b9a\u8bad\u7ec3\u96c6\u3001\u9a8c\u8bc1\u96c6\u8def\u5f84\u548c\u7c7b\u522b\u540d\u79f0\u3002\u4f8b\u5982\uff1a\n<pre><code>train: .\/dataset\/train\/images\r\nval: .\/dataset\/val\/images\r\nnc: 2  # \u7c7b\u522b\u6570\u91cf\r\nnames: ['cat', 'dog']  # \u7c7b\u522b\u540d\u79f0\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u52a0\u8f7d\u6a21\u578b\u5e76\u8bad\u7ec3<\/strong><br \/>\n\u4f7f\u7528 Python \u811a\u672c\u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\u5e76\u5f00\u59cb\u8bad\u7ec3\uff1a<\/p>\n<pre><code>from ultralytics import YOLO\r\nmodel = YOLO('yolov12s.pt')  # \u53ef\u9009 n\/s\/m\/l\/x \u6a21\u578b\r\nresults = model.train(data='path\/to\/data.yaml', epochs=250, imgsz=640)\r\n<\/code><\/pre>\n<ul>\n<li><code>epochs<\/code>\uff1a\u8bad\u7ec3\u8f6e\u6570\uff0c\u5efa\u8bae 250 \u6b21\u4ee5\u4e0a\u4ee5\u83b7\u5f97\u66f4\u597d\u6548\u679c\u3002<\/li>\n<li><code>imgsz<\/code>\uff1a\u8f93\u5165\u56fe\u50cf\u5c3a\u5bf8\uff0c\u9ed8\u8ba4 640&#215;640\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u67e5\u770b\u8bad\u7ec3\u7ed3\u679c<\/strong><br \/>\n\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u7ed3\u679c\u4fdd\u5b58\u5728\u00a0<code>runs\/detect\/train<\/code>\u00a0\u6587\u4ef6\u5939\uff0c\u5305\u62ec\u6a21\u578b\u6743\u91cd\uff08<code>best.pt<\/code>\uff09\u548c\u6df7\u6dc6\u77e9\u9635\u7b49\u3002\u8fd0\u884c\u4ee5\u4e0b\u4ee3\u7801\u67e5\u770b\u6df7\u6dc6\u77e9\u9635\uff1a<\/p>\n<pre><code>from IPython.display import Image\r\nImage(filename='runs\/detect\/train\/confusion_matrix.png', width=600)\r\n<\/code><\/pre>\n<\/li>\n<\/ol>\n<h4>\u63a8\u7406\u4e0e\u6d4b\u8bd5<\/h4>\n<p>\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u53ef\u7528\u4e8e\u56fe\u50cf\u6216\u89c6\u9891\u7684\u76ee\u6807\u68c0\u6d4b\uff1a<\/p>\n<ol>\n<li><strong>\u5355\u5f20\u56fe\u50cf\u68c0\u6d4b<\/strong>\n<pre><code>model = YOLO('path\/to\/best.pt')\r\nresults = model('test.jpg')\r\nresults.show()  # \u663e\u793a\u68c0\u6d4b\u7ed3\u679c\r\nresults.save()  # \u4fdd\u5b58\u7ed3\u679c\u5230 runs\/detect\/predict\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u89c6\u9891\u68c0\u6d4b<\/strong><br \/>\n\u4f7f\u7528\u547d\u4ee4\u884c\u5904\u7406\u89c6\u9891\u6587\u4ef6\uff1a<\/p>\n<pre><code>python app.py --source 'video.mp4' --model 'path\/to\/best.pt'\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u6027\u80fd\u8bc4\u4f30<\/strong><br \/>\n\u5bf9\u9a8c\u8bc1\u96c6\u8fdb\u884c\u8bc4\u4f30\uff0c\u83b7\u53d6 mAP \u7b49\u6307\u6807\uff1a<\/p>\n<pre><code>results = model.val(data='path\/to\/data.yaml')\r\nprint(results.box.map)  # \u8f93\u51fa mAP@0.5:0.95\r\n<\/code><\/pre>\n<\/li>\n<\/ol>\n<h4>\u6a21\u578b\u5bfc\u51fa<\/h4>\n<p>\u5c06\u6a21\u578b\u5bfc\u51fa\u4e3a\u751f\u4ea7\u73af\u5883\u53ef\u7528\u7684\u683c\u5f0f\uff1a<\/p>\n<pre><code>model.export(format='onnx', half=True)  # \u5bfc\u51fa\u4e3a ONNX\uff0c\u652f\u6301 FP16 \u52a0\u901f\r\n<\/code><\/pre>\n<p>\u5bfc\u51fa\u7684\u6a21\u578b\u53ef\u90e8\u7f72\u5230\u8fb9\u7f18\u8bbe\u5907\u6216\u670d\u52a1\u5668\u4e0a\u3002<\/p>\n<h3>\u7279\u8272\u529f\u80fd\u64cd\u4f5c<\/h3>\n<ul>\n<li><strong>\u6ce8\u610f\u529b\u673a\u5236\u4f18\u5316<\/strong><br \/>\nYOLOv12 \u7684\u201c\u533a\u57df\u6ce8\u610f\u529b\u201d\u6a21\u5757\u65e0\u9700\u624b\u52a8\u914d\u7f6e\uff0c\u4f1a\u81ea\u52a8\u5728\u8bad\u7ec3\u548c\u63a8\u7406\u4e2d\u4f18\u5316\u7279\u5f81\u63d0\u53d6\uff0c\u63d0\u5347\u5c0f\u76ee\u6807\u68c0\u6d4b\u80fd\u529b\u3002\u7528\u6237\u53ea\u9700\u9009\u62e9\u5408\u9002\u7684\u6a21\u578b\u89c4\u6a21\uff08\u5982 Nano \u7528\u4e8e\u4f4e\u529f\u8017\u8bbe\u5907\uff09\uff0c\u5373\u53ef\u4eab\u53d7\u8fd9\u4e00\u7279\u6027\u5e26\u6765\u7684\u7cbe\u5ea6\u63d0\u5347\u3002<\/li>\n<li><strong>\u5b9e\u65f6\u68c0\u6d4b<\/strong><br \/>\n\u5728\u652f\u6301 CUDA \u7684 GPU \u4e0a\u8fd0\u884c\u65f6\uff0c\u63a8\u7406\u901f\u5ea6\u6781\u5feb\u3002\u4f8b\u5982\uff0c\u4f7f\u7528 T4 GPU \u8fd0\u884c YOLOv12-N \u6a21\u578b\uff0c\u5355\u5f20\u56fe\u50cf\u68c0\u6d4b\u4ec5\u9700 1.64ms\u3002\u7528\u6237\u53ef\u901a\u8fc7\u76d1\u7763\u5de5\u5177\uff08supervision\uff09\u5b9e\u65f6\u53ef\u89c6\u5316\u68c0\u6d4b\u6846\u548c\u7f6e\u4fe1\u5ea6\uff1a<\/p>\n<pre><code>results = model('image.jpg')\r\nresults.plot()  # \u663e\u793a\u5e26\u6807\u6ce8\u7684\u56fe\u50cf\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u591a\u573a\u666f\u9002\u914d<\/strong><br \/>\n\u901a\u8fc7\u8c03\u6574\u6a21\u578b\u89c4\u6a21\u548c\u8bad\u7ec3\u6570\u636e\uff0cYOLOv12 \u53ef\u8f7b\u677e\u9002\u914d\u4e0d\u540c\u4efb\u52a1\u3002\u4f8b\u5982\uff0c\u5728\u76d1\u63a7\u7cfb\u7edf\u4e2d\u68c0\u6d4b\u884c\u4eba\uff0c\u6216\u5728\u81ea\u52a8\u9a7e\u9a76\u4e2d\u8bc6\u522b\u8f66\u8f86\u548c\u4ea4\u901a\u6807\u5fd7\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>YOLOv12 \u662f\u7531 GitHub \u7528\u6237 sunsmarterjie \u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u4e13\u6ce8\u4e8e\u5b9e\u65f6\u76ee\u6807\u68c0\u6d4b\u6280\u672f\u3002\u8be5\u9879\u76ee\u57fa\u4e8e YOLO\uff08You Only Look Once\uff09\u7cfb\u5217\u6846\u67b6\uff0c\u5f15\u5165\u6ce8\u610f\u529b\u673a\u5236\u4f18\u5316\u4f20\u7edf\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08CNN\uff09\u7684\u6027\u80fd\uff0c\u4e0d\u4ec5&#8230;<\/p>\n","protected":false},"author":1,"featured_media":61900,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230,378],"class_list":["post-26478","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-aikaiyuanxiangmu","tag-shijuemubiaojiance"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/26478","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/comments?post=26478"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/26478\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media\/61900"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media?parent=26478"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=26478"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=26478"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}