devbox
我家自动化
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

YOLOv10环境搭建、模型预测和ONNX推理_yolov10安装

作者：我家自动化 | 2024-06-09 05:34:02

赞

踩

yolov10安装

目录

1、开源代码、模型下载

2、环境配置

3、模型预测

4、onnxruntime测试

1、开源代码、模型下载

代码下载链接：https://github.com/THU-MIG/yolov10

模型下载：

YOLOv10-N：https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10n.pt

YOLOv10-S：https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10s.pt

YOLOv10-M：https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10m.pt

YOLOv10-B：https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10b.pt

YOLOv10-L：https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10l.pt

YOLOv10-X：https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10x.pt

2、环境配置

打开Anaconda3终端，进入base环境，创建新环境

  1. conda create -n yolov10 python=3.9
  2. conda activate yolov10
  3. #cd到yolov10的目录下
  4. pip install -r requirements.txt -i  https://pypi.tuna.tsinghua.edu.cn/simple/
  5. pip install -e .

3、模型预测

安装完成之后，我们简单执行下推理命令测试下效果，默认读取yolov10-main/ultralytics/assets文件夹下的所有图像：

yolo predict model=yolov10s.pt

或者使用脚本

  1.  
  2.  
  3. from ultralytics import YOLOv10
  4. import glob
  5. import os
  6. import numpy as np
  7. import cv2
  8.  
  9. classes = {
  10.     0: 'person', 1: 'bicycle', 2: 'car', 3: 'motorcycle', 4: 'airplane', 5: 'bus',
  11.     6: 'train', 7: 'truck', 8: 'boat', 9: 'traffic light', 10: 'fire hydrant',
  12.     11: 'stop sign', 12: 'parking meter', 13: 'bench', 14: 'bird', 15: 'cat',
  13.     16: 'dog', 17: 'horse', 18: 'sheep', 19: 'cow', 20: 'elephant', 21: 'bear',
  14.     22: 'zebra', 23: 'giraffe', 24: 'backpack', 25: 'umbrella', 26: 'handbag',
  15.     27: 'tie', 28: 'suitcase', 29: 'frisbee', 30: 'skis', 31: 'snowboard',
  16.     32: 'sports ball', 33: 'kite', 34: 'baseball bat', 35: 'baseball glove',
  17.     36: 'skateboard', 37: 'surfboard', 38: 'tennis racket', 39: 'bottle',
  18.     40: 'wine glass', 41: 'cup', 42: 'fork', 43: 'knife', 44: 'spoon', 45: 'bowl',
  19.     46: 'banana', 47: 'apple', 48: 'sandwich', 49: 'orange', 50: 'broccoli',
  20.     51: 'carrot', 52: 'hot dog', 53: 'pizza', 54: 'donut', 55: 'cake',
  21.     56: 'chair', 57: 'couch', 58: 'potted plant', 59: 'bed', 60: 'dining table',
  22.     61: 'toilet', 62: 'tv', 63: 'laptop', 64: 'mouse', 65: 'remote', 66: 'keyboard',
  23.     67: 'cell phone', 68: 'microwave', 69: 'oven', 70: 'toaster', 71: 'sink',
  24.     72: 'refrigerator', 73: 'book', 74: 'clock', 75: 'vase', 76: 'scissors',
  25.     77: 'teddy bear', 78: 'hair drier', 79: 'toothbrush'
  26. }
  27. class Colors:
  28.     """Ultralytics color palette https://ultralytics.com/."""
  29.  
  30.     def __init__(self):
  31.         """Initialize colors as hex = matplotlib.colors.TABLEAU_COLORS.values()."""
  32.         hexs = ('FF3838', 'FF9D97', 'FF701F', 'FFB21D', 'CFD231', '48F90A', '92CC17', '3DDB86', '1A9334', '00D4BB',
  33.                 '2C99A8', '00C2FF', '344593', '6473FF', '0018EC', '8438FF', '520085', 'CB38FF', 'FF95C8', 'FF37C7')
  34.         self.palette = [self.hex2rgb(f'#{c}') for c in hexs]
  35.         # print(self.palette)
  36.         self.n = len(self.palette)
  37.  
  38.     def __call__(self, i, bgr=False):
  39.         """Converts hex color codes to rgb values."""
  40.         c = self.palette[int(i) % self.n]
  41.         return (c[2], c[1], c[0]) if bgr else c
  42.  
  43.     @staticmethod
  44.     def hex2rgb(h):  # rgb order (PIL)
  45.         return tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))
  46.  
  47.  
  48. colors = Colors()  # create instance for 'from utils.plots import colors'
  49.  
  50. imgpath = r'D:\Yolov10\yolov10-main\yolov10-detect\test2'
  51. modelpath = r'D:\Yolov10\yolov10-main\yolov10-detect\yolov10s.pt'
  52. save_dir = imgpath + '_Rst'
  53. os.makedirs(save_dir,exist_ok=True)
  54. model = YOLOv10(modelpath)
  55.  
  56. imgs = glob.glob(os.path.join(imgpath,'*.jpg'))
  57. for img in imgs:
  58.     imgname = img.split('\\')[-1]
  59.     frame = cv2.imread(img)
  60.     results = model.predict(img)[0]
  61.     # results = model(img)
  62.  
  63.     for box in results.boxes:
  64.         # print(box)
  65.         xyxy = box.xyxy.squeeze().tolist()
  66.         x1, y1, x2, y2 = int(xyxy[0]), int(xyxy[1]), int(xyxy[2]), int(xyxy[3])
  67.         c, conf = int(box.cls), float(box.conf)
  68.         name = classes[c]
  69.         color = colors(c, True)
  70.         cv2.rectangle(frame, (int(xyxy[0]), int(xyxy[1])), (int(xyxy[2]), int(xyxy[3])), color, thickness=2, lineType=cv2.LINE_AA)
  71.         cv2.putText(frame, f"{name}: {conf:.2f}", (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 1, color,
  72.                     2)
  73.     # cv2.imshow('image', frame)
  74.     # cv2.waitKey(0)
  75.     cv2.imwrite(save_dir+'\\'+imgname,frame)
  76.

运行结果如下：

4、onnxruntime测试

（1）onnx模型转换

yolo export model=yolov10s.pt format=onnx opset=13 simplify

 运行后会在文件yolov10s.pt存放路径下生成一个的yolov10s.onnxONNX模型文件

可以通过这个网站Netron查看导出的节点信息：

（2）模型推理

通过 Ultralytics 框架测试下能否正常推理：

yolo predict model=yolov10s.onnx

或者使用推理脚本

  1. import glob
  2. import os
  3. import cv2
  4. import numpy as np
  5. import onnxruntime as ort
  6.  
  7.  
  8. classes = {
  9.     0: 'person', 1: 'bicycle', 2: 'car', 3: 'motorcycle', 4: 'airplane', 5: 'bus',
  10.     6: 'train', 7: 'truck', 8: 'boat', 9: 'traffic light', 10: 'fire hydrant',
  11.     11: 'stop sign', 12: 'parking meter', 13: 'bench', 14: 'bird', 15: 'cat',
  12.     16: 'dog', 17: 'horse', 18: 'sheep', 19: 'cow', 20: 'elephant', 21: 'bear',
  13.     22: 'zebra', 23: 'giraffe', 24: 'backpack', 25: 'umbrella', 26: 'handbag',
  14.     27: 'tie', 28: 'suitcase', 29: 'frisbee', 30: 'skis', 31: 'snowboard',
  15.     32: 'sports ball', 33: 'kite', 34: 'baseball bat', 35: 'baseball glove',
  16.     36: 'skateboard', 37: 'surfboard', 38: 'tennis racket', 39: 'bottle',
  17.     40: 'wine glass', 41: 'cup', 42: 'fork', 43: 'knife', 44: 'spoon', 45: 'bowl',
  18.     46: 'banana', 47: 'apple', 48: 'sandwich', 49: 'orange', 50: 'broccoli',
  19.     51: 'carrot', 52: 'hot dog', 53: 'pizza', 54: 'donut', 55: 'cake',
  20.     56: 'chair', 57: 'couch', 58: 'potted plant', 59: 'bed', 60: 'dining table',
  21.     61: 'toilet', 62: 'tv', 63: 'laptop', 64: 'mouse', 65: 'remote', 66: 'keyboard',
  22.     67: 'cell phone', 68: 'microwave', 69: 'oven', 70: 'toaster', 71: 'sink',
  23.     72: 'refrigerator', 73: 'book', 74: 'clock', 75: 'vase', 76: 'scissors',
  24.     77: 'teddy bear', 78: 'hair drier', 79: 'toothbrush'
  25. }
  26. class Colors:
  27.     """Ultralytics color palette https://ultralytics.com/."""
  28.  
  29.     def __init__(self):
  30.         """Initialize colors as hex = matplotlib.colors.TABLEAU_COLORS.values()."""
  31.         hexs = ('FF3838', 'FF9D97', 'FF701F', 'FFB21D', 'CFD231', '48F90A', '92CC17', '3DDB86', '1A9334', '00D4BB',
  32.                 '2C99A8', '00C2FF', '344593', '6473FF', '0018EC', '8438FF', '520085', 'CB38FF', 'FF95C8', 'FF37C7')
  33.         self.palette = [self.hex2rgb(f'#{c}') for c in hexs]
  34.         # print(self.palette)
  35.         self.n = len(self.palette)
  36.  
  37.     def __call__(self, i, bgr=False):
  38.         """Converts hex color codes to rgb values."""
  39.         c = self.palette[int(i) % self.n]
  40.         return (c[2], c[1], c[0]) if bgr else c
  41.  
  42.     @staticmethod
  43.     def hex2rgb(h):  # rgb order (PIL)
  44.         return tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))
  45.  
  46.  
  47. colors = Colors()  # create instance for 'from utils.plots import colors'
  48.  
  49.  
  50. def letterbox(
  51.         im,
  52.         new_shape,
  53.         color=(114, 114, 114),
  54.         auto=False,
  55.         scaleFill=False,
  56.         scaleup=True,
  57.         stride=32,
  58. ):
  59.     """
  60.     Resize and pad image while meeting stride-multiple constraints
  61.     Returns:
  62.         im (array): (height, width, 3)
  63.         ratio (array): [w_ratio, h_ratio]
  64.         (dw, dh) (array): [w_padding h_padding]
  65.     """
  66.     shape = im.shape[:2]  # current shape [height, width]
  67.     if isinstance(new_shape, int):  # [h_rect, w_rect]
  68.         new_shape = (new_shape, new_shape)
  69.  
  70.     # Scale ratio (new / old)
  71.     r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
  72.     if not scaleup:  # only scale down, do not scale up (for better val mAP)
  73.         r = min(r, 1.0)
  74.  
  75.     # Compute padding
  76.     ratio = r, r  # wh ratios
  77.     new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))  # w h
  78.     dw, dh = (
  79.         new_shape[1] - new_unpad[0],
  80.         new_shape[0] - new_unpad[1],
  81.     )  # wh padding
  82.  
  83.     if auto:  # minimum rectangle
  84.         dw, dh = np.mod(dw, stride), np.mod(dh, stride)  # wh padding
  85.     elif scaleFill:  # stretch
  86.         dw, dh = 0.0, 0.0
  87.         new_unpad = (new_shape[1], new_shape[0])  # [w h]
  88.         ratio = (
  89.             new_shape[1] / shape[1],
  90.             new_shape[0] / shape[0],
  91.         )  # [w_ratio, h_ratio]
  92.  
  93.     dw /= 2  # divide padding into 2 sides
  94.     dh /= 2
  95.     if shape[::-1] != new_unpad:  # resize
  96.         im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
  97.     top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
  98.     left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
  99.     im = cv2.copyMakeBorder(
  100.         im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color
  101.     )
  102.     return im, ratio, (dw, dh)
  103.  
  104.  
  105. def rescale_coords(boxes, image_shape, input_shape):
  106.     image_height, image_width = image_shape
  107.     input_height, input_width = input_shape
  108.  
  109.     scale = min(input_width / image_width, input_height / image_height)
  110.  
  111.     pad_w = (input_width - image_width * scale) / 2
  112.     pad_h = (input_height - image_height * scale) / 2
  113.  
  114.     boxes[:, [0, 2]] = (boxes[:, [0, 2]] - pad_w) / scale
  115.     boxes[:, [1, 3]] = (boxes[:, [1, 3]] - pad_h) / scale
  116.  
  117.     boxes[:, [0, 2]] = np.clip(boxes[:, [0, 2]], 0, image_width)
  118.     boxes[:, [1, 3]] = np.clip(boxes[:, [1, 3]], 0, image_height)
  119.  
  120.     return boxes.astype(int)
  121.  
  122.  
  123. def preprocess(image, input_shape):
  124.     # Resize
  125.     input_img = letterbox(image, input_shape)[0]
  126.     # Transpose
  127.     input_img = input_img[..., ::-1].transpose(2, 0, 1)
  128.     # Expand
  129.     input_img = input_img[np.newaxis, :, :, :].astype(np.float32)
  130.     # Contiguous
  131.     input_img = np.ascontiguousarray(input_img)
  132.     # Norm
  133.     blob = input_img / 255.0
  134.     return blob
  135.  
  136.  
  137. def postprocess(outs, conf_thres, image_shape, input_shape):
  138.     # Filtered by conf
  139.     outs = outs[outs[:, 4] >= conf_thres]
  140.  
  141.     # Extract
  142.     boxes = outs[:, :4]
  143.     scores = outs[:, -2]
  144.     labels = outs[:, -1].astype(int)
  145.  
  146.     # Rescale
  147.     boxes = rescale_coords(boxes, image_shape, input_shape)
  148.  
  149.     return boxes, scores, labels
  150.  
  151.  
  152. def main():
  153.     conf_thres = 0.25
  154.     input_shape = (640, 640)
  155.     image_path = r'D:\Yolov10\yolov10-main\ultralytics\assets'
  156.     save_path = image_path + '_Rst'
  157.     os.makedirs(save_path,exist_ok=True)
  158.     model_path = r'D:\Yolov10\yolov10-main\yolov10-detect\yolov10s.onnx'
  159.  
  160.     ort_model = ort.InferenceSession(model_path)
  161.     imgs = glob.glob(os.path.join(image_path,'*.jpg'))
  162.     imgs.sort()
  163.     for img in imgs:
  164.         imgname = img.split('\\')[-1]
  165.         # Preprocess
  166.         im0 = cv2.imread(img)
  167.         image_shape = im0.shape[:2]
  168.         blob = preprocess(im0, input_shape)
  169.  
  170.         # Inference
  171.         outs = ort_model.run(None, {'images': blob})[0][0]
  172.  
  173.         # Postprocess
  174.         boxes, scores, labels = postprocess(outs, conf_thres, image_shape, input_shape)
  175.  
  176.         # 保存结果
  177.         for label, score, box in zip(labels, scores, boxes):
  178.             label_text = f'{classes[label]}: {score:.2f}'
  179.             color = colors(label,True)
  180.             cv2.rectangle(im0, (box[0], box[1]), (box[2], box[3]), color, thickness=2, lineType=cv2.LINE_AA)
  181.             cv2.putText(im0, label_text, (box[0], box[1] - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
  182.  
  183.         # cv2.imshow('image', im0)
  184.         # cv2.waitKey(0)
  185.         cv2.imwrite(save_path+'\\'+imgname, im0)
  186.  
  187.  
  188. if __name__ == '__main__':
  189.     main()

参考：YOLOv10 正式发布！原理、部署、应用一站式齐全-CSDN博客

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/我家自动化/article/detail/692823
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号