{"id":28200,"date":"2025-04-19T23:26:48","date_gmt":"2025-04-19T15:26:48","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/28200.html"},"modified":"2025-04-19T23:26:48","modified_gmt":"2025-04-19T15:26:48","slug":"atlas800%e6%98%87%e8%85%be%e6%9c%8d%e5%8a%a1%e5%99%a8%ef%bc%88%e5%9e%8b%e5%8f%b7%ef%bc%9a3000%ef%bc%89-yolo%e5%85%a8%e7%b3%bb%e5%88%97npu%e6%8e%a8%e7%90%86%e3%80%90%e6%a3%80%e6%b5%8b%e3%80%91","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/28200.html","title":{"rendered":"Atlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09"},"content":{"rendered":"

\u670d\u52a1\u5668\u914d\u7f6e\u5982\u4e0b：<\/h3>\n

CPU\/NPU：\u9cb2\u9e4f CPU（ARM64）+A300I pro\u63a8\u7406\u5361 \u7cfb\u7edf：Kylin V10 SP1\u3010\u4e0b\u8f7d\u94fe\u63a5\u3011\u3010\u5b89\u88c5\u94fe\u63a5\u3011 \u9a71\u52a8\u4e0e\u56fa\u4ef6\u7248\u672c\u7248\u672c： Ascend-hdk-310p-npu-driver_23.0.1_linux-aarch64.run\u3010\u4e0b\u8f7d\u94fe\u63a5\u3011 Ascend-hdk-310p-npu-firmware_7.1.0.4.220.run\u3010\u4e0b\u8f7d\u94fe\u63a5\u3011 MCU\u7248\u672c：Ascend-hdk-310p-mcu_23.2.3\u3010\u4e0b\u8f7d\u94fe\u63a5\u3011 CANN\u5f00\u53d1\u5957\u4ef6：\u7248\u672c7.0.1\u3010Toolkit\u4e0b\u8f7d\u94fe\u63a5\u3011\u3010Kernels\u4e0b\u8f7d\u94fe\u63a5\u3011<\/p>\n

\u6d4b\u8bd5om\u6a21\u578b\u73af\u5883\u5982\u4e0b：<\/h3>\n

Python：\u7248\u672c3.8.11 \u63a8\u7406\u5de5\u5177：ais_bench \u6d4b\u8bd5YOLO\u7cfb\u5217：v5\/6\/7\/8\/9\/10\/11<\/p>\n

\u4e13\u680f\u5176\u4ed6\u6587\u7ae0： Atlas800\u6607\u817e\u670d\u52a1\u5668（\u578b\u53f7：3000）\u2014\u9a71\u52a8\u4e0e\u56fa\u4ef6\u5b89\u88c5（\u4e00） Atlas800\u6607\u817e\u670d\u52a1\u5668（\u578b\u53f7：3000）\u2014CANN\u5b89\u88c5（\u4e8c） Atlas800\u6607\u817e\u670d\u52a1\u5668（\u578b\u53f7：3000）\u2014YOLO\u5168\u7cfb\u5217om\u6a21\u578b\u8f6c\u6362\u6d4b\u8bd5（\u4e09） Atlas800\u6607\u817e\u670d\u52a1\u5668（\u578b\u53f7：3000）\u2014AIPP\u52a0\u901f\u524d\u5904\u7406（\u56db） Atlas800\u6607\u817e\u670d\u52a1\u5668（\u578b\u53f7：3000）\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011（\u4e94） Atlas800\u6607\u817e\u670d\u52a1\u5668（\u578b\u53f7：3000）\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u5b9e\u4f8b\u5206\u5272\u3011（\u516d） Atlas800\u6607\u817e\u670d\u52a1\u5668（\u578b\u53f7：3000）\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u5173\u952e\u70b9\u3011（\u4e03） Atlas800\u6607\u817e\u670d\u52a1\u5668（\u578b\u53f7：3000）\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u8ddf\u8e2a\u3011（\u516b）<\/p>\n

\u5168\u90e8\u4ee3\u7801github：https:\/\/github.com\/Bigtuo\/NPU-ais_bench<\/p>\n

1 \u57fa\u7840\u73af\u5883\u5b89\u88c5<\/h3>\n

\u8be6\u60c5\u89c1\u7b2c（\u4e09）\u7ae0\u73af\u5883\u5b89\u88c5：https:\/\/blog.csdn.net\/weixin_45679938\/article\/details\/142966255<\/p>\n

2 ais_bench\u7f16\u8bd1\u5b89\u88c5<\/h3>\n

\u6ce8\u610f：\u76ee\u524dais_bench\u5de5\u5177\u53ea\u652f\u6301\u5355\u4e2ainput\u7684\u5e26\u6709\u52a8\u6001AIPP\u914d\u7f6e\u7684\u6a21\u578b，\u53ea\u652f\u6301\u9759\u6001shape\u3001\u52a8\u6001batch\u3001\u52a8\u6001\u5bbd\u9ad8\u4e09\u79cd\u573a\u666f，\u4e0d\u652f\u6301\u52a8\u6001shape\u573a\u666f\u3002 \u53c2\u8003\u94fe\u63a5：https:\/\/gitee.com\/ascend\/tools\/tree\/master\/ais-bench_workload\/tool\/ais_bench<\/p>\n

2.1 \u5b89\u88c5aclruntime\u5305<\/h4>\n

\u5728\u5b89\u88c5\u73af\u5883\u6267\u884c\u5982\u4e0b\u547d\u4ee4\u5b89\u88c5aclruntime\u5305： \u8bf4\u660e：\u82e5\u4e3a\u8986\u76d6\u5b89\u88c5，\u8bf7\u589e\u52a0\u201c\u2013force-reinstall\u201d\u53c2\u6570\u5f3a\u5236\u5b89\u88c5.<\/p>\n

pip3 install –<\/span>v 'git+https:\/\/gitee.com\/ascend\/tools.git#egg=aclruntime&subdirectory=ais-bench_workload\/tool\/ais_bench\/backend'<\/span> –<\/span>i https:<\/span>\/\/<\/span>pypi.<\/span>tuna.<\/span>tsinghua.<\/span>edu.<\/span>cn\/<\/span>simple<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

2.2 \u5b89\u88c5ais_bench\u63a8\u7406\u7a0b\u5e8f\u5305<\/h4>\n

\u5728\u5b89\u88c5\u73af\u5883\u6267\u884c\u5982\u4e0b\u547d\u4ee4\u5b89\u88c5ais_bench\u63a8\u7406\u7a0b\u5e8f\u5305：<\/p>\n

pip3 install –<\/span>v 'git+https:\/\/gitee.com\/ascend\/tools.git#egg=ais_bench&subdirectory=ais-bench_workload\/tool\/ais_bench'<\/span> –<\/span>i https:<\/span>\/\/<\/span>pypi.<\/span>tuna.<\/span>tsinghua.<\/span>edu.<\/span>cn\/<\/span>simple<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \u5378\u8f7d\u548c\u66f4\u65b0\u3010\u5ffd\u7565\u3011：<\/p>\n

# \u5378\u8f7daclruntime<\/span>
\npip3 uninstall aclruntime
\n# \u5378\u8f7dais_bench\u63a8\u7406\u7a0b\u5e8f<\/span>
\npip3 uninstall ais_bench<\/p>\n

3 \u88f8\u4ee3\u7801\u63a8\u7406\u6d4b\u8bd5<\/h3>\n

# 1.\u8fdb\u5165\u8fd0\u884c\u73af\u5883yolo\u3010\u666e\u901a\u7528\u6237\u3011<\/span>
\nconda activate yolo
\n# 2.\u6fc0\u6d3batc\u3010atc –help\u6d4b\u8bd5\u662f\u5426\u53ef\u884c\u3011<\/span>
\nsource ~<\/span>\/<\/span>.<\/span>bashrc<\/p>\n

\u6ce8\u610f：ais_bench\u8c03\u7528\u548c\u4f7f\u7528\u65b9\u5f0f\u4e0eonnx-runtime\u51e0\u4e4e\u4e00\u81f4，\u56e0\u6b64\u53ef\u53c2\u8003\u8fdb\u884c\u64b0\u5199\u811a\u672c！<\/p>\n

\u4ee3\u7801\u903b\u8f91\u5982\u4e0b： \u4e0b\u9762\u4ee3\u7801\u6574\u4e2a\u5904\u7406\u8fc7\u7a0b\u4e3b\u8981\u5305\u62ec：\u9884\u5904\u7406\u2014>\u63a8\u7406\u2014>\u540e\u5904\u7406\u2014>\u753b\u56fe\u3002 \u5047\u8bbe\u56fe\u50cfresize\u4e3a640\u00d7640， \u524d\u5904\u7406\u8f93\u51fa\u7ed3\u679c\u7ef4\u5ea6：(1, 3, 640, 640)； YOLOv5\/6\/7\u63a8\u7406\u8f93\u51fa\u7ed3\u679c\u7ef4\u5ea6：(1, 8400\u00d73, 85)，\u5176\u4e2d85\u8868\u793a4\u4e2abox\u5750\u6807\u4fe1\u606f+\u7f6e\u4fe1\u5ea6\u5206\u6570+80\u4e2a\u7c7b\u522b\u6982\u7387，8400\u00d73\u8868\u793a(80\u00d780+40\u00d740+20\u00d720)\u00d73，\u4e0d\u540c\u4e8ev8\u4e0ev9\u91c7\u7528\u7c7b\u522b\u91cc\u9762\u6700\u5927\u7684\u6982\u7387\u4f5c\u4e3a\u7f6e\u4fe1\u5ea6score； YOLOv8\/9\/11\u63a8\u7406\u8f93\u51fa\u7ed3\u679c\u7ef4\u5ea6：(1, 84, 8400)，\u5176\u4e2d84\u8868\u793a4\u4e2abox\u5750\u6807\u4fe1\u606f+80\u4e2a\u7c7b\u522b\u6982\u7387，8400\u8868\u793a80\u00d780+40\u00d740+20\u00d720； YOLOv10\u63a8\u7406\u8f93\u51fa\u7ed3\u679c\u7ef4\u5ea6：(1, 300, 6)，\u5176\u4e2d300\u662f\u9ed8\u8ba4\u8f93\u51fa\u6570\u91cf，\u65e0nms\u64cd\u4f5c，\u9608\u503c\u8fc7\u6ee4\u5373\u53ef，6\u662f4\u4e2abox\u5750\u6807\u4fe1\u606f+\u7f6e\u4fe1\u5ea6\u5206\u6570+\u7c7b\u522b\u3002 \u540e\u5904\u7406\u8f93\u51fa\u7ed3\u679c\u7ef4\u5ea6：(5, 6)，\u5176\u4e2d\u7b2c\u4e00\u4e2a5\u8868\u793a\u56febus.jpg\u68c0\u51fa5\u4e2a\u76ee\u6807，\u7b2c\u4e8c\u4e2a\u7ef4\u5ea66\u8868\u793a(x1, y1, x2, y2, conf, cls)\u3002<\/p>\n

\u5b8c\u6574\u4ee3\u7801\u5982\u4e0b： \u65b0\u5efaYOLO_ais_bench_det_aipp.py，\u5185\u5bb9\u5982\u4e0b：<\/p>\n

import<\/span> argparse
\nimport<\/span> time
\nimport<\/span> cv2
\nimport<\/span> numpy as<\/span> np
\nimport<\/span> os<\/p>\n

from<\/span> ais_bench.<\/span>infer.<\/span>interface import<\/span> InferSession<\/p>\n

class<\/span> YOLO<\/span>:<\/span>
\n """YOLO object detection model class for handling inference"""<\/span><\/p>\n

def<\/span> __init__<\/span>(<\/span>self,<\/span> om_model,<\/span> imgsz=<\/span>(<\/span>640<\/span>,<\/span> 640<\/span>)<\/span>,<\/span> device_id=<\/span>0<\/span>,<\/span> model_ndtype=<\/span>np.<\/span>single,<\/span> mode=<\/span>"static"<\/span>,<\/span> postprocess_type=<\/span>"v8"<\/span>,<\/span> aipp=<\/span>False<\/span>)<\/span>:<\/span>
\n """
\n Initialization.<\/p>\n

Args:
\n om_model (str): Path to the om model.
\n """<\/span><\/p>\n

# \u6784\u5efaais_bench\u63a8\u7406\u5f15\u64ce<\/span>
\n self.<\/span>session =<\/span> InferSession(<\/span>device_id=<\/span>device_id,<\/span> model_path=<\/span>om_model)<\/span><\/p>\n

# Numpy dtype: support both FP32(np.single) and FP16(np.half) om model<\/span>
\n self.<\/span>ndtype =<\/span> model_ndtype
\n self.<\/span>mode =<\/span> mode
\n self.<\/span>postprocess_type =<\/span> postprocess_type
\n self.<\/span>aipp =<\/span> aipp <\/p>\n

self.<\/span>model_height,<\/span> self.<\/span>model_width =<\/span> imgsz[<\/span>0<\/span>]<\/span>,<\/span> imgsz[<\/span>1<\/span>]<\/span> # \u56fe\u50cfresize\u5927\u5c0f<\/span><\/p>\n

def<\/span> __call__<\/span>(<\/span>self,<\/span> im0,<\/span> conf_threshold=<\/span>0.4<\/span>,<\/span> iou_threshold=<\/span>0.45<\/span>)<\/span>:<\/span>
\n """
\n The whole pipeline: pre-process -> inference -> post-process.<\/p>\n

Args:
\n im0 (Numpy.ndarray): original input image.
\n conf_threshold (float): confidence threshold for filtering predictions.
\n iou_threshold (float): iou threshold for NMS.<\/p>\n

Returns:
\n boxes (List): list of bounding boxes.
\n """<\/span>
\n # \u524d\u5904\u7406Pre-process<\/span>
\n t1 =<\/span> time.<\/span>time(<\/span>)<\/span>
\n im,<\/span> ratio,<\/span> (<\/span>pad_w,<\/span> pad_h)<\/span> =<\/span> self.<\/span>preprocess(<\/span>im0)<\/span>
\n pre_time =<\/span> round<\/span>(<\/span>time.<\/span>time(<\/span>)<\/span> –<\/span> t1,<\/span> 3<\/span>)<\/span><\/p>\n

# \u63a8\u7406 inference<\/span>
\n t2 =<\/span> time.<\/span>time(<\/span>)<\/span>
\n preds =<\/span> self.<\/span>session.<\/span>infer(<\/span>[<\/span>im]<\/span>,<\/span> mode=<\/span>self.<\/span>mode)<\/span>[<\/span>0<\/span>]<\/span> # mode\u6709\u52a8\u6001"dymshape"\u548c\u9759\u6001"static"\u7b49<\/span>
\n det_time =<\/span> round<\/span>(<\/span>time.<\/span>time(<\/span>)<\/span> –<\/span> t2,<\/span> 3<\/span>)<\/span><\/p>\n

# \u540e\u5904\u7406Post-process<\/span>
\n t3 =<\/span> time.<\/span>time(<\/span>)<\/span>
\n if<\/span> self.<\/span>postprocess_type ==<\/span> "v5"<\/span>:<\/span>
\n boxes =<\/span> self.<\/span>postprocess_v5(<\/span>preds,<\/span>
\n im0=<\/span>im0,<\/span>
\n ratio=<\/span>ratio,<\/span>
\n pad_w=<\/span>pad_w,<\/span>
\n pad_h=<\/span>pad_h,<\/span>
\n conf_threshold=<\/span>conf_threshold,<\/span>
\n iou_threshold=<\/span>iou_threshold,<\/span>
\n )<\/span><\/p>\n

elif<\/span> self.<\/span>postprocess_type ==<\/span> "v8"<\/span>:<\/span>
\n boxes =<\/span> self.<\/span>postprocess_v8(<\/span>preds,<\/span>
\n im0=<\/span>im0,<\/span>
\n ratio=<\/span>ratio,<\/span>
\n pad_w=<\/span>pad_w,<\/span>
\n pad_h=<\/span>pad_h,<\/span>
\n conf_threshold=<\/span>conf_threshold,<\/span>
\n iou_threshold=<\/span>iou_threshold,<\/span>
\n )<\/span><\/p>\n

elif<\/span> self.<\/span>postprocess_type ==<\/span> "v10"<\/span>:<\/span>
\n boxes =<\/span> self.<\/span>postprocess_v10(<\/span>preds,<\/span>
\n im0=<\/span>im0,<\/span>
\n ratio=<\/span>ratio,<\/span>
\n pad_w=<\/span>pad_w,<\/span>
\n pad_h=<\/span>pad_h,<\/span>
\n conf_threshold=<\/span>conf_threshold
\n )<\/span><\/p>\n

else<\/span>:<\/span>
\n boxes =<\/span> [<\/span>]<\/span><\/p>\n

post_time =<\/span> round<\/span>(<\/span>time.<\/span>time(<\/span>)<\/span> –<\/span> t3,<\/span> 3<\/span>)<\/span><\/p>\n

return<\/span> boxes,<\/span> (<\/span>pre_time,<\/span> det_time,<\/span> post_time)<\/span><\/p>\n

# \u524d\u5904\u7406，\u5305\u62ec：resize, pad, \u5176\u4e2dHWC to CHW，BGR to RGB，\u5f52\u4e00\u5316，\u589e\u52a0\u7ef4\u5ea6CHW -> BCHW\u53ef\u9009\u62e9\u662f\u5426\u5f00\u542fAIPP\u52a0\u901f\u5904\u7406<\/span>
\n def<\/span> preprocess<\/span>(<\/span>self,<\/span> img)<\/span>:<\/span>
\n """
\n Pre-processes the input image.<\/p>\n

Args:
\n img (Numpy.ndarray): image about to be processed.<\/p>\n

Returns:
\n img_process (Numpy.ndarray): image preprocessed for inference.
\n ratio (tuple): width, height ratios in letterbox.
\n pad_w (float): width padding in letterbox.
\n pad_h (float): height padding in letterbox.
\n """<\/span>
\n # Resize and pad input image using letterbox() (Borrowed from Ultralytics)<\/span>
\n shape =<\/span> img.<\/span>shape[<\/span>:<\/span>2<\/span>]<\/span> # original image shape<\/span>
\n new_shape =<\/span> (<\/span>self.<\/span>model_height,<\/span> self.<\/span>model_width)<\/span>
\n r =<\/span> min<\/span>(<\/span>new_shape[<\/span>0<\/span>]<\/span> \/<\/span> shape[<\/span>0<\/span>]<\/span>,<\/span> new_shape[<\/span>1<\/span>]<\/span> \/<\/span> shape[<\/span>1<\/span>]<\/span>)<\/span>
\n ratio =<\/span> r,<\/span> r
\n new_unpad =<\/span> int<\/span>(<\/span>round<\/span>(<\/span>shape[<\/span>1<\/span>]<\/span> *<\/span> r)<\/span>)<\/span>,<\/span> int<\/span>(<\/span>round<\/span>(<\/span>shape[<\/span>0<\/span>]<\/span> *<\/span> r)<\/span>)<\/span>
\n pad_w,<\/span> pad_h =<\/span> (<\/span>new_shape[<\/span>1<\/span>]<\/span> –<\/span> new_unpad[<\/span>0<\/span>]<\/span>)<\/span> \/<\/span> 2<\/span>,<\/span> (<\/span>new_shape[<\/span>0<\/span>]<\/span> –<\/span> new_unpad[<\/span>1<\/span>]<\/span>)<\/span> \/<\/span> 2<\/span> # wh padding<\/span>
\n if<\/span> shape[<\/span>:<\/span>:<\/span>–<\/span>1<\/span>]<\/span> !=<\/span> new_unpad:<\/span> # resize<\/span>
\n img =<\/span> cv2.<\/span>resize(<\/span>img,<\/span> new_unpad,<\/span> interpolation=<\/span>cv2.<\/span>INTER_LINEAR)<\/span><\/p>\n

top,<\/span> bottom =<\/span> int<\/span>(<\/span>round<\/span>(<\/span>pad_h –<\/span> 0.1<\/span>)<\/span>)<\/span>,<\/span> int<\/span>(<\/span>round<\/span>(<\/span>pad_h +<\/span> 0.1<\/span>)<\/span>)<\/span>
\n left,<\/span> right =<\/span> int<\/span>(<\/span>round<\/span>(<\/span>pad_w –<\/span> 0.1<\/span>)<\/span>)<\/span>,<\/span> int<\/span>(<\/span>round<\/span>(<\/span>pad_w +<\/span> 0.1<\/span>)<\/span>)<\/span>
\n img =<\/span> cv2.<\/span>copyMakeBorder(<\/span>img,<\/span> top,<\/span> bottom,<\/span> left,<\/span> right,<\/span> cv2.<\/span>BORDER_CONSTANT,<\/span> value=<\/span>(<\/span>114<\/span>,<\/span> 114<\/span>,<\/span> 114<\/span>)<\/span>)<\/span> # \u586b\u5145<\/span><\/p>\n

# \u662f\u5426\u5f00\u542faipp\u52a0\u901f\u9884\u5904\u7406，\u9700atc\u4e2d\u5b8c\u6210<\/span>
\n if<\/span> self.<\/span>aipp:<\/span>
\n return<\/span> img,<\/span> ratio,<\/span> (<\/span>pad_w,<\/span> pad_h)<\/span><\/p>\n

# Transforms: HWC to CHW -> BGR to RGB -> div(255) -> contiguous -> add axis(optional)<\/span>
\n img =<\/span> np.<\/span>ascontiguousarray(<\/span>np.<\/span>einsum(<\/span>'HWC->CHW'<\/span>,<\/span> img)<\/span>[<\/span>:<\/span>:<\/span>–<\/span>1<\/span>]<\/span>,<\/span> dtype=<\/span>self.<\/span>ndtype)<\/span> \/<\/span> 255.0<\/span>
\n img_process =<\/span> img[<\/span>None<\/span>]<\/span> if<\/span> len<\/span>(<\/span>img.<\/span>shape)<\/span> ==<\/span> 3<\/span> else<\/span> img
\n return<\/span> img_process,<\/span> ratio,<\/span> (<\/span>pad_w,<\/span> pad_h)<\/span><\/p>\n

# YOLOv5\/6\/7\u901a\u7528\u540e\u5904\u7406，\u5305\u62ec：\u9608\u503c\u8fc7\u6ee4\u4e0eNMS<\/span>
\n def<\/span> postprocess_v5<\/span>(<\/span>self,<\/span> preds,<\/span> im0,<\/span> ratio,<\/span> pad_w,<\/span> pad_h,<\/span> conf_threshold,<\/span> iou_threshold)<\/span>:<\/span>
\n """
\n Post-process the prediction.<\/p>\n

Args:
\n preds (Numpy.ndarray): predictions come from ort.session.run().
\n im0 (Numpy.ndarray): [h, w, c] original input image.
\n ratio (tuple): width, height ratios in letterbox.
\n pad_w (float): width padding in letterbox.
\n pad_h (float): height padding in letterbox.
\n conf_threshold (float): conf threshold.
\n iou_threshold (float): iou threshold.<\/p>\n

Returns:
\n boxes (List): list of bounding boxes.
\n """<\/span>
\n # (Batch_size, Num_anchors, xywh_score_conf_cls), v5\u548cv6\u7684[…, 4]\u662f\u7f6e\u4fe1\u5ea6\u5206\u6570，v8v9\u91c7\u7528\u7c7b\u522b\u91cc\u9762\u6700\u5927\u7684\u6982\u7387\u4f5c\u4e3a\u7f6e\u4fe1\u5ea6score<\/span>
\n x =<\/span> preds # outputs: predictions (1, 8400*3, 85)<\/span><\/p>\n

# Predictions filtering by conf-threshold<\/span>
\n x =<\/span> x[<\/span>x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> 4<\/span>]<\/span> ><\/span> conf_threshold]<\/span><\/p>\n

# Create a new matrix which merge these(box, score, cls) into one<\/span>
\n # For more details about `numpy.c_()`: https:\/\/numpy.org\/doc\/1.26\/reference\/generated\/numpy.c_.html<\/span>
\n x =<\/span> np.<\/span>c_[<\/span>x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>4<\/span>]<\/span>,<\/span> x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> 4<\/span>]<\/span>,<\/span> np.<\/span>argmax(<\/span>x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> 5<\/span>:<\/span>]<\/span>,<\/span> axis=<\/span>–<\/span>1<\/span>)<\/span>]<\/span><\/p>\n

# NMS filtering<\/span>
\n # \u7ecf\u8fc7NMS\u540e\u7684\u503c, np.array([[x, y, w, h, conf, cls], …]), shape=(-1, 4 + 1 + 1)<\/span>
\n x =<\/span> x[<\/span>cv2.<\/span>dnn.<\/span>NMSBoxes(<\/span>x[<\/span>:<\/span>,<\/span> :<\/span>4<\/span>]<\/span>,<\/span> x[<\/span>:<\/span>,<\/span> 4<\/span>]<\/span>,<\/span> conf_threshold,<\/span> iou_threshold)<\/span>]<\/span><\/p>\n

# \u91cd\u65b0\u7f29\u653e\u8fb9\u754c\u6846，\u4e3a\u753b\u56fe\u505a\u51c6\u5907<\/span>
\n if<\/span> len<\/span>(<\/span>x)<\/span> ><\/span> 0<\/span>:<\/span>
\n # Bounding boxes format change: cxcywh -> xyxy<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 1<\/span>]<\/span>]<\/span> -=<\/span> x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>2<\/span>,<\/span> 3<\/span>]<\/span>]<\/span> \/<\/span> 2<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>2<\/span>,<\/span> 3<\/span>]<\/span>]<\/span> +=<\/span> x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 1<\/span>]<\/span>]<\/span><\/p>\n

# Rescales bounding boxes from model shape(model_height, model_width) to the shape of original image<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>4<\/span>]<\/span> -=<\/span> [<\/span>pad_w,<\/span> pad_h,<\/span> pad_w,<\/span> pad_h]<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>4<\/span>]<\/span> \/=<\/span> min<\/span>(<\/span>ratio)<\/span><\/p>\n

# Bounding boxes boundary clamp<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 2<\/span>]<\/span>]<\/span> =<\/span> x[<\/span>:<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 2<\/span>]<\/span>]<\/span>.<\/span>clip(<\/span>0<\/span>,<\/span> im0.<\/span>shape[<\/span>1<\/span>]<\/span>)<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>1<\/span>,<\/span> 3<\/span>]<\/span>]<\/span> =<\/span> x[<\/span>:<\/span>,<\/span> [<\/span>1<\/span>,<\/span> 3<\/span>]<\/span>]<\/span>.<\/span>clip(<\/span>0<\/span>,<\/span> im0.<\/span>shape[<\/span>0<\/span>]<\/span>)<\/span><\/p>\n

return<\/span> x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>6<\/span>]<\/span> # boxes<\/span>
\n else<\/span>:<\/span>
\n return<\/span> [<\/span>]<\/span><\/p>\n

# YOLOv8\/9\/11\u901a\u7528\u540e\u5904\u7406，\u5305\u62ec：\u9608\u503c\u8fc7\u6ee4\u4e0eNMS<\/span>
\n def<\/span> postprocess_v8<\/span>(<\/span>self,<\/span> preds,<\/span> im0,<\/span> ratio,<\/span> pad_w,<\/span> pad_h,<\/span> conf_threshold,<\/span> iou_threshold)<\/span>:<\/span>
\n """
\n Post-process the prediction.<\/p>\n

Args:
\n preds (Numpy.ndarray): predictions come from ort.session.run().
\n im0 (Numpy.ndarray): [h, w, c] original input image.
\n ratio (tuple): width, height ratios in letterbox.
\n pad_w (float): width padding in letterbox.
\n pad_h (float): height padding in letterbox.
\n conf_threshold (float): conf threshold.
\n iou_threshold (float): iou threshold.<\/p>\n

Returns:
\n boxes (List): list of bounding boxes.
\n """<\/span>
\n x =<\/span> preds # outputs: predictions (1, 84, 8400)<\/span>
\n # Transpose the first output: (Batch_size, xywh_conf_cls, Num_anchors) -> (Batch_size, Num_anchors, xywh_conf_cls)<\/span>
\n x =<\/span> np.<\/span>einsum(<\/span>'bcn->bnc'<\/span>,<\/span> x)<\/span> # (1, 8400, 84)<\/span><\/p>\n

# Predictions filtering by conf-threshold<\/span>
\n x =<\/span> x[<\/span>np.<\/span>amax(<\/span>x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> 4<\/span>:<\/span>]<\/span>,<\/span> axis=<\/span>–<\/span>1<\/span>)<\/span> ><\/span> conf_threshold]<\/span><\/p>\n

# Create a new matrix which merge these(box, score, cls) into one<\/span>
\n # For more details about `numpy.c_()`: https:\/\/numpy.org\/doc\/1.26\/reference\/generated\/numpy.c_.html<\/span>
\n x =<\/span> np.<\/span>c_[<\/span>x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>4<\/span>]<\/span>,<\/span> np.<\/span>amax(<\/span>x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> 4<\/span>:<\/span>]<\/span>,<\/span> axis=<\/span>–<\/span>1<\/span>)<\/span>,<\/span> np.<\/span>argmax(<\/span>x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> 4<\/span>:<\/span>]<\/span>,<\/span> axis=<\/span>–<\/span>1<\/span>)<\/span>]<\/span><\/p>\n

# NMS filtering<\/span>
\n # \u7ecf\u8fc7NMS\u540e\u7684\u503c, np.array([[x, y, w, h, conf, cls], …]), shape=(-1, 4 + 1 + 1)<\/span>
\n x =<\/span> x[<\/span>cv2.<\/span>dnn.<\/span>NMSBoxes(<\/span>x[<\/span>:<\/span>,<\/span> :<\/span>4<\/span>]<\/span>,<\/span> x[<\/span>:<\/span>,<\/span> 4<\/span>]<\/span>,<\/span> conf_threshold,<\/span> iou_threshold)<\/span>]<\/span><\/p>\n

# \u91cd\u65b0\u7f29\u653e\u8fb9\u754c\u6846，\u4e3a\u753b\u56fe\u505a\u51c6\u5907<\/span>
\n if<\/span> len<\/span>(<\/span>x)<\/span> ><\/span> 0<\/span>:<\/span>
\n # Bounding boxes format change: cxcywh -> xyxy<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 1<\/span>]<\/span>]<\/span> -=<\/span> x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>2<\/span>,<\/span> 3<\/span>]<\/span>]<\/span> \/<\/span> 2<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>2<\/span>,<\/span> 3<\/span>]<\/span>]<\/span> +=<\/span> x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 1<\/span>]<\/span>]<\/span><\/p>\n

# Rescales bounding boxes from model shape(model_height, model_width) to the shape of original image<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>4<\/span>]<\/span> -=<\/span> [<\/span>pad_w,<\/span> pad_h,<\/span> pad_w,<\/span> pad_h]<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>4<\/span>]<\/span> \/=<\/span> min<\/span>(<\/span>ratio)<\/span><\/p>\n

# Bounding boxes boundary clamp<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 2<\/span>]<\/span>]<\/span> =<\/span> x[<\/span>:<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 2<\/span>]<\/span>]<\/span>.<\/span>clip(<\/span>0<\/span>,<\/span> im0.<\/span>shape[<\/span>1<\/span>]<\/span>)<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>1<\/span>,<\/span> 3<\/span>]<\/span>]<\/span> =<\/span> x[<\/span>:<\/span>,<\/span> [<\/span>1<\/span>,<\/span> 3<\/span>]<\/span>]<\/span>.<\/span>clip(<\/span>0<\/span>,<\/span> im0.<\/span>shape[<\/span>0<\/span>]<\/span>)<\/span><\/p>\n

return<\/span> x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>6<\/span>]<\/span> # boxes<\/span>
\n else<\/span>:<\/span>
\n return<\/span> [<\/span>]<\/span><\/p>\n

# YOLOv10\u540e\u5904\u7406，\u5305\u62ec：\u9608\u503c\u8fc7\u6ee4-\u65e0NMS<\/span>
\n def<\/span> postprocess_v10<\/span>(<\/span>self,<\/span> preds,<\/span> im0,<\/span> ratio,<\/span> pad_w,<\/span> pad_h,<\/span> conf_threshold)<\/span>:<\/span><\/p>\n

x =<\/span> preds # outputs: predictions (1, 300, 6) -> (xyxy_conf_cls)<\/span><\/p>\n

# Predictions filtering by conf-threshold<\/span>
\n x =<\/span> x[<\/span>x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> 4<\/span>]<\/span> ><\/span> conf_threshold]<\/span><\/p>\n

# \u91cd\u65b0\u7f29\u653e\u8fb9\u754c\u6846，\u4e3a\u753b\u56fe\u505a\u51c6\u5907<\/span>
\n if<\/span> len<\/span>(<\/span>x)<\/span> ><\/span> 0<\/span>:<\/span><\/p>\n

# Rescales bounding boxes from model shape(model_height, model_width) to the shape of original image<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>4<\/span>]<\/span> -=<\/span> [<\/span>pad_w,<\/span> pad_h,<\/span> pad_w,<\/span> pad_h]<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> :<\/span>4<\/span>]<\/span> \/=<\/span> min<\/span>(<\/span>ratio)<\/span><\/p>\n

# Bounding boxes boundary clamp<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 2<\/span>]<\/span>]<\/span> =<\/span> x[<\/span>:<\/span>,<\/span> [<\/span>0<\/span>,<\/span> 2<\/span>]<\/span>]<\/span>.<\/span>clip(<\/span>0<\/span>,<\/span> im0.<\/span>shape[<\/span>1<\/span>]<\/span>)<\/span>
\n x[<\/span>.<\/span>.<\/span>.<\/span>,<\/span> [<\/span>1<\/span>,<\/span> 3<\/span>]<\/span>]<\/span> =<\/span> x[<\/span>:<\/span>,<\/span> [<\/span>1<\/span>,<\/span> 3<\/span>]<\/span>]<\/span>.<\/span>clip(<\/span>0<\/span>,<\/span> im0.<\/span>shape[<\/span>0<\/span>]<\/span>)<\/span><\/p>\n

return<\/span> x # boxes<\/span>
\n else<\/span>:<\/span>
\n return<\/span> [<\/span>]<\/span><\/p>\n

if<\/span> __name__ ==<\/span> '__main__'<\/span>:<\/span>
\n # Create an argument parser to handle command-line arguments<\/span>
\n parser =<\/span> argparse.<\/span>ArgumentParser(<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–det_model'<\/span>,<\/span> type<\/span>=<\/span>str<\/span>,<\/span> default=<\/span>r"yolov8s.om"<\/span>,<\/span> help<\/span>=<\/span>'Path to OM model'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–source'<\/span>,<\/span> type<\/span>=<\/span>str<\/span>,<\/span> default=<\/span>r'images'<\/span>,<\/span> help<\/span>=<\/span>'Path to input image'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–out_path'<\/span>,<\/span> type<\/span>=<\/span>str<\/span>,<\/span> default=<\/span>r'results'<\/span>,<\/span> help<\/span>=<\/span>'\u7ed3\u679c\u4fdd\u5b58\u6587\u4ef6\u5939'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–imgsz_det'<\/span>,<\/span> type<\/span>=<\/span>tuple<\/span>,<\/span> default=<\/span>(<\/span>640<\/span>,<\/span> 640<\/span>)<\/span>,<\/span> help<\/span>=<\/span>'Image input size'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–classes'<\/span>,<\/span> type<\/span>=<\/span>list<\/span>,<\/span> default=<\/span>[<\/span>'person'<\/span>,<\/span> 'bicycle'<\/span>,<\/span> 'car'<\/span>,<\/span> 'motorcycle'<\/span>,<\/span> 'airplane'<\/span>,<\/span> 'bus'<\/span>,<\/span> 'train'<\/span>,<\/span> 'truck'<\/span>,<\/span> 'boat'<\/span>,<\/span> 'traffic light'<\/span>,<\/span>
\n 'fire hydrant'<\/span>,<\/span> 'stop sign'<\/span>,<\/span> 'parking meter'<\/span>,<\/span> 'bench'<\/span>,<\/span> 'bird'<\/span>,<\/span> 'cat'<\/span>,<\/span> 'dog'<\/span>,<\/span> 'horse'<\/span>,<\/span> 'sheep'<\/span>,<\/span> 'cow'<\/span>,<\/span>
\n 'elephant'<\/span>,<\/span> 'bear'<\/span>,<\/span> 'zebra'<\/span>,<\/span> 'giraffe'<\/span>,<\/span> 'backpack'<\/span>,<\/span> 'umbrella'<\/span>,<\/span> 'handbag'<\/span>,<\/span> 'tie'<\/span>,<\/span> 'suitcase'<\/span>,<\/span> 'frisbee'<\/span>,<\/span>
\n 'skis'<\/span>,<\/span> 'snowboard'<\/span>,<\/span> 'sports ball'<\/span>,<\/span> 'kite'<\/span>,<\/span> 'baseball bat'<\/span>,<\/span> 'baseball glove'<\/span>,<\/span> 'skateboard'<\/span>,<\/span> 'surfboard'<\/span>,<\/span>
\n 'tennis racket'<\/span>,<\/span> 'bottle'<\/span>,<\/span> 'wine glass'<\/span>,<\/span> 'cup'<\/span>,<\/span> 'fork'<\/span>,<\/span> 'knife'<\/span>,<\/span> 'spoon'<\/span>,<\/span> 'bowl'<\/span>,<\/span> 'banana'<\/span>,<\/span> 'apple'<\/span>,<\/span> 'sandwich'<\/span>,<\/span>
\n 'orange'<\/span>,<\/span> 'broccoli'<\/span>,<\/span> 'carrot'<\/span>,<\/span> 'hot dog'<\/span>,<\/span> 'pizza'<\/span>,<\/span> 'donut'<\/span>,<\/span> 'cake'<\/span>,<\/span> 'chair'<\/span>,<\/span> 'couch'<\/span>,<\/span> 'potted plant'<\/span>,<\/span> 'bed'<\/span>,<\/span>
\n 'dining table'<\/span>,<\/span> 'toilet'<\/span>,<\/span> 'tv'<\/span>,<\/span> 'laptop'<\/span>,<\/span> 'mouse'<\/span>,<\/span> 'remote'<\/span>,<\/span> 'keyboard'<\/span>,<\/span> 'cell phone'<\/span>,<\/span> 'microwave'<\/span>,<\/span> 'oven'<\/span>,<\/span>
\n 'toaster'<\/span>,<\/span> 'sink'<\/span>,<\/span> 'refrigerator'<\/span>,<\/span> 'book'<\/span>,<\/span> 'clock'<\/span>,<\/span> 'vase'<\/span>,<\/span> 'scissors'<\/span>,<\/span> 'teddy bear'<\/span>,<\/span> 'hair drier'<\/span>,<\/span> 'toothbrush'<\/span>]<\/span>,<\/span> help<\/span>=<\/span>'\u7c7b\u522b'<\/span>)<\/span><\/p>\n

parser.<\/span>add_argument(<\/span>'–conf'<\/span>,<\/span> type<\/span>=<\/span>float<\/span>,<\/span> default=<\/span>0.25<\/span>,<\/span> help<\/span>=<\/span>'Confidence threshold'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–iou'<\/span>,<\/span> type<\/span>=<\/span>float<\/span>,<\/span> default=<\/span>0.6<\/span>,<\/span> help<\/span>=<\/span>'NMS IoU threshold'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–device_id'<\/span>,<\/span> type<\/span>=<\/span>int<\/span>,<\/span> default=<\/span>0<\/span>,<\/span> help<\/span>=<\/span>'device id'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–mode'<\/span>,<\/span> default=<\/span>'static'<\/span>,<\/span> help<\/span>=<\/span>'om\u662f\u52a8\u6001dymshape\u6216\u9759\u6001static'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–model_ndtype'<\/span>,<\/span> default=<\/span>np.<\/span>single,<\/span> help<\/span>=<\/span>'om\u662ffp32\u6216fp16'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–postprocess_type'<\/span>,<\/span> type<\/span>=<\/span>str<\/span>,<\/span> default=<\/span>'v8'<\/span>,<\/span> help<\/span>=<\/span>'\u540e\u5904\u7406\u65b9\u5f0f, \u5bf9\u5e94v5\/v8\/v10\u4e09\u79cd\u540e\u5904\u7406'<\/span>)<\/span>
\n parser.<\/span>add_argument(<\/span>'–aipp'<\/span>,<\/span> default=<\/span>False<\/span>,<\/span> action=<\/span>'store_true'<\/span>,<\/span> help<\/span>=<\/span>'\u662f\u5426\u5f00\u542faipp\u52a0\u901fYOLO\u9884\u5904\u7406, \u9700atc\u4e2d\u5b8c\u6210om\u96c6\u6210'<\/span>)<\/span>
\n args =<\/span> parser.<\/span>parse_args(<\/span>)<\/span><\/p>\n

# \u521b\u5efa\u7ed3\u679c\u4fdd\u5b58\u6587\u4ef6\u5939<\/span>
\n if<\/span> not<\/span> os.<\/span>path.<\/span>exists(<\/span>args.<\/span>out_path)<\/span>:<\/span>
\n os.<\/span>mkdir(<\/span>args.<\/span>out_path)<\/span><\/p>\n

print<\/span>(<\/span>'\u5f00\u59cb\u8fd0\u884c：'<\/span>)<\/span>
\n # Build model<\/span>
\n det_model =<\/span> YOLO(<\/span>args.<\/span>det_model,<\/span> args.<\/span>imgsz_det,<\/span> args.<\/span>device_id,<\/span> args.<\/span>model_ndtype,<\/span> args.<\/span>mode,<\/span> args.<\/span>postprocess_type,<\/span> args.<\/span>aipp)<\/span>
\n color_palette =<\/span> np.<\/span>random.<\/span>uniform(<\/span>0<\/span>,<\/span> 255<\/span>,<\/span> size=<\/span>(<\/span>len<\/span>(<\/span>args.<\/span>classes)<\/span>,<\/span> 3<\/span>)<\/span>)<\/span> # \u4e3a\u6bcf\u4e2a\u7c7b\u522b\u751f\u6210\u8c03\u8272\u677f<\/span><\/p>\n

for<\/span> i,<\/span> img_name in<\/span> enumerate<\/span>(<\/span>os.<\/span>listdir(<\/span>args.<\/span>source)<\/span>)<\/span>:<\/span>
\n try<\/span>:<\/span>
\n t1 =<\/span> time.<\/span>time(<\/span>)<\/span>
\n # Read image by OpenCV<\/span>
\n img =<\/span> cv2.<\/span>imread(<\/span>os.<\/span>path.<\/span>join(<\/span>args.<\/span>source,<\/span> img_name)<\/span>)<\/span><\/p>\n

# \u68c0\u6d4bInference<\/span>
\n boxes,<\/span> (<\/span>pre_time,<\/span> det_time,<\/span> post_time)<\/span> =<\/span> det_model(<\/span>img,<\/span> conf_threshold=<\/span>args.<\/span>conf,<\/span> iou_threshold=<\/span>args.<\/span>iou)<\/span>
\n print<\/span>(<\/span>'{}\/{} ==>\u603b\u8017\u65f6\u95f4: {:.3f}s, \u5176\u4e2d, \u9884\u5904\u7406: {:.3f}s, \u63a8\u7406: {:.3f}s, \u540e\u5904\u7406: {:.3f}s, \u8bc6\u522b{}\u4e2a\u76ee\u6807'<\/span>.<\/span>format<\/span>(<\/span>i+<\/span>1<\/span>,<\/span> len<\/span>(<\/span>os.<\/span>listdir(<\/span>args.<\/span>source)<\/span>)<\/span>,<\/span> time.<\/span>time(<\/span>)<\/span> –<\/span> t1,<\/span> pre_time,<\/span> det_time,<\/span> post_time,<\/span> len<\/span>(<\/span>boxes)<\/span>)<\/span>)<\/span><\/p>\n

# Draw rectangles<\/span>
\n for<\/span> (<\/span>*<\/span>box,<\/span> conf,<\/span> cls_)<\/span> in<\/span> boxes:<\/span>
\n cv2.<\/span>rectangle(<\/span>img,<\/span> (<\/span>int<\/span>(<\/span>box[<\/span>0<\/span>]<\/span>)<\/span>,<\/span> int<\/span>(<\/span>box[<\/span>1<\/span>]<\/span>)<\/span>)<\/span>,<\/span> (<\/span>int<\/span>(<\/span>box[<\/span>2<\/span>]<\/span>)<\/span>,<\/span> int<\/span>(<\/span>box[<\/span>3<\/span>]<\/span>)<\/span>)<\/span>,<\/span>
\n color_palette[<\/span>int<\/span>(<\/span>cls_)<\/span>]<\/span>,<\/span> 2<\/span>,<\/span> cv2.<\/span>LINE_AA)<\/span>
\n cv2.<\/span>putText(<\/span>img,<\/span> f'<\/span>{<\/span>args.<\/span>classes[<\/span>int<\/span>(<\/span>cls_)<\/span>]<\/span>}<\/span><\/span>: <\/span>{<\/span>conf:<\/span>.3f<\/span>}<\/span><\/span>'<\/span><\/span>,<\/span> (<\/span>int<\/span>(<\/span>box[<\/span>0<\/span>]<\/span>)<\/span>,<\/span> int<\/span>(<\/span>box[<\/span>1<\/span>]<\/span> –<\/span> 9<\/span>)<\/span>)<\/span>,<\/span>
\n cv2.<\/span>FONT_HERSHEY_SIMPLEX,<\/span> 1<\/span>,<\/span> (<\/span>0<\/span>,<\/span> 0<\/span>,<\/span> 255<\/span>)<\/span>,<\/span> 2<\/span>,<\/span> cv2.<\/span>LINE_AA)<\/span><\/p>\n

cv2.<\/span>imwrite(<\/span>os.<\/span>path.<\/span>join(<\/span>args.<\/span>out_path,<\/span> img_name)<\/span>,<\/span> img)<\/span><\/p>\n

except<\/span> Exception as<\/span> e:<\/span>
\n print<\/span>(<\/span>e)<\/span> <\/p>\n

\u68c0\u6d4b\u7ed3\u679c\u53ef\u89c6\u5316\u5982\u4e0b，\u6548\u679c\u4e0eGPU\u4e0a\u63a8\u7406\u51e0\u4e4e\u4e00\u81f4： \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

4 \u63a8\u7406\u8017\u65f6<\/h3>\n

YOLO\u5404\u7cfb\u5217\u63a8\u7406\u8017\u65f6（640*640）\u5982\u4e0b： YOLOv5s：8-9ms YOLOv7-tiny：7-8ms YOLOv7：14ms YOLOv8s：6ms YOLOv9s：12ms YOLOv10s：6ms YOLOv11s：8ms \u9884\u5904\u7406\u8017\u65f6（bus.jpg）：12ms \u540e\u5904\u7406\u8017\u65f6：\u9664YOLOv10\u51e0\u4e4e\u65e0\u8017\u65f6\u5916，\u5176\u4f591-2ms\u3002 \u6ce8\u610f，\u4e0a\u8ff0\u8017\u65f6\u672a\u4f7f\u7528AIPP\u8fdb\u884c\u524d\u5904\u7406\u52a0\u901f，\u5982YOLOv8s\u52a0\u901f\u540e\u524d\u5904\u7406+\u63a8\u7406\u5927\u7ea66-7ms\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb2.4k\u6b21\uff0c\u70b9\u8d5e53\u6b21\uff0c\u6536\u85cf25\u6b21\u3002\u76ee\u524dais_bench\u5de5\u5177\u53ea\u652f\u6301\u5355\u4e2ainput\u7684\u5e26\u6709\u52a8\u6001AIPP\u914d\u7f6e\u7684\u6a21\u578b\uff0c\u53ea\u652f\u6301\u9759\u6001shape\u3001\u52a8\u6001batch\u3001\u52a8\u6001\u5bbd\u9ad8\u4e09\u79cd\u573a\u666f\uff0c\u4e0d\u652f\u6301\u52a8\u6001shape\u573a\u666f\u3002\uff1aais_bench\u8c03\u7528\u548c\u4f7f\u7528\u65b9\u5f0f\u4e0eonnx-runtime\u51e0\u4e4e\u4e00\u81f4\uff0c\u56e0\u6b64\u53ef\u53c2\u8003\u8fdb\u884c\u64b0\u5199\u811a\u672c\uff01\u6ce8\u610f\uff0c\u4e0a\u8ff0\u8017\u65f6\u672a\u4f7f\u7528AIPP\u8fdb\u884c\u524d\u5904\u7406\u52a0\u901f\uff0c\u5982YOLOv8s\u52a0\u901f\u540e\u524d\u5904\u7406+\u63a8\u7406\u5927\u7ea66-7ms\u3002\u8bf4\u660e\uff1a\u82e5\u4e3a\u8986\u76d6\u5b89\u88c5\uff0c\u8bf7\u589e\u52a0\u201c\u2013force-reinstall\u201d\u53c2\u6570\u5f3a\u5236\u5b89\u88c5.\u540e\u5904\u7406\u8017\u65f6\uff1a\u9664YOLOv10\u51e0\u4e4e\u65e0\u8017\u65f6\u5916\uff0c\u5176\u4f591-2ms\u3002\u5f00\u53d1\u5957\u4ef6\uff1a\u7248\u672c7.0.1\u3010_\u534e\u4e3a\u6607\u817egpu\u670d\u52a1\u5668\u6d4b\u8bd5<\/p>\n","protected":false},"author":2,"featured_media":28197,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[1235,2257,1234,1233,69,2256],"topic":[],"class_list":["post-28200","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-a300i-pro","tag-asi_bench","tag-atlas800","tag-atlas800om","tag-npu","tag-yolov8"],"yoast_head":"\nAtlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/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:\/\/www.wsisp.com\/helps\/28200.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Atlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb2.4k\u6b21\uff0c\u70b9\u8d5e53\u6b21\uff0c\u6536\u85cf25\u6b21\u3002\u76ee\u524dais_bench\u5de5\u5177\u53ea\u652f\u6301\u5355\u4e2ainput\u7684\u5e26\u6709\u52a8\u6001AIPP\u914d\u7f6e\u7684\u6a21\u578b\uff0c\u53ea\u652f\u6301\u9759\u6001shape\u3001\u52a8\u6001batch\u3001\u52a8\u6001\u5bbd\u9ad8\u4e09\u79cd\u573a\u666f\uff0c\u4e0d\u652f\u6301\u52a8\u6001shape\u573a\u666f\u3002\uff1aais_bench\u8c03\u7528\u548c\u4f7f\u7528\u65b9\u5f0f\u4e0eonnx-runtime\u51e0\u4e4e\u4e00\u81f4\uff0c\u56e0\u6b64\u53ef\u53c2\u8003\u8fdb\u884c\u64b0\u5199\u811a\u672c\uff01\u6ce8\u610f\uff0c\u4e0a\u8ff0\u8017\u65f6\u672a\u4f7f\u7528AIPP\u8fdb\u884c\u524d\u5904\u7406\u52a0\u901f\uff0c\u5982YOLOv8s\u52a0\u901f\u540e\u524d\u5904\u7406+\u63a8\u7406\u5927\u7ea66-7ms\u3002\u8bf4\u660e\uff1a\u82e5\u4e3a\u8986\u76d6\u5b89\u88c5\uff0c\u8bf7\u589e\u52a0\u201c\u2013force-reinstall\u201d\u53c2\u6570\u5f3a\u5236\u5b89\u88c5.\u540e\u5904\u7406\u8017\u65f6\uff1a\u9664YOLOv10\u51e0\u4e4e\u65e0\u8017\u65f6\u5916\uff0c\u5176\u4f591-2ms\u3002\u5f00\u53d1\u5957\u4ef6\uff1a\u7248\u672c7.0.1\u3010_\u534e\u4e3a\u6607\u817egpu\u670d\u52a1\u5668\u6d4b\u8bd5\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/28200.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2025-04-19T15:26:48+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/04\/20250419152647-6803c0b769a14.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"10 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/28200.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/28200.html\",\"name\":\"Atlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2025-04-19T15:26:48+00:00\",\"dateModified\":\"2025-04-19T15:26:48+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/28200.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/28200.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/28200.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Atlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Atlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","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:\/\/www.wsisp.com\/helps\/28200.html","og_locale":"zh_CN","og_type":"article","og_title":"Atlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb2.4k\u6b21\uff0c\u70b9\u8d5e53\u6b21\uff0c\u6536\u85cf25\u6b21\u3002\u76ee\u524dais_bench\u5de5\u5177\u53ea\u652f\u6301\u5355\u4e2ainput\u7684\u5e26\u6709\u52a8\u6001AIPP\u914d\u7f6e\u7684\u6a21\u578b\uff0c\u53ea\u652f\u6301\u9759\u6001shape\u3001\u52a8\u6001batch\u3001\u52a8\u6001\u5bbd\u9ad8\u4e09\u79cd\u573a\u666f\uff0c\u4e0d\u652f\u6301\u52a8\u6001shape\u573a\u666f\u3002\uff1aais_bench\u8c03\u7528\u548c\u4f7f\u7528\u65b9\u5f0f\u4e0eonnx-runtime\u51e0\u4e4e\u4e00\u81f4\uff0c\u56e0\u6b64\u53ef\u53c2\u8003\u8fdb\u884c\u64b0\u5199\u811a\u672c\uff01\u6ce8\u610f\uff0c\u4e0a\u8ff0\u8017\u65f6\u672a\u4f7f\u7528AIPP\u8fdb\u884c\u524d\u5904\u7406\u52a0\u901f\uff0c\u5982YOLOv8s\u52a0\u901f\u540e\u524d\u5904\u7406+\u63a8\u7406\u5927\u7ea66-7ms\u3002\u8bf4\u660e\uff1a\u82e5\u4e3a\u8986\u76d6\u5b89\u88c5\uff0c\u8bf7\u589e\u52a0\u201c\u2013force-reinstall\u201d\u53c2\u6570\u5f3a\u5236\u5b89\u88c5.\u540e\u5904\u7406\u8017\u65f6\uff1a\u9664YOLOv10\u51e0\u4e4e\u65e0\u8017\u65f6\u5916\uff0c\u5176\u4f591-2ms\u3002\u5f00\u53d1\u5957\u4ef6\uff1a\u7248\u672c7.0.1\u3010_\u534e\u4e3a\u6607\u817egpu\u670d\u52a1\u5668\u6d4b\u8bd5","og_url":"https:\/\/www.wsisp.com\/helps\/28200.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2025-04-19T15:26:48+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/04\/20250419152647-6803c0b769a14.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"10 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/28200.html","url":"https:\/\/www.wsisp.com\/helps\/28200.html","name":"Atlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2025-04-19T15:26:48+00:00","dateModified":"2025-04-19T15:26:48+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/28200.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/28200.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/28200.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"Atlas800\u6607\u817e\u670d\u52a1\u5668\uff08\u578b\u53f7\uff1a3000\uff09\u2014YOLO\u5168\u7cfb\u5217NPU\u63a8\u7406\u3010\u68c0\u6d4b\u3011\uff08\u4e94\uff09"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/28200","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=28200"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/28200\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/28197"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=28200"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=28200"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=28200"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=28200"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}