实时目标检测使用Yolo模型不起作用。

huangapple
go评论136阅读模式
英文:

Real time object detection with Yolo model not working

问题

我已经训练了一个自定义的YOLO模型，用于检测棋盘上的正方形槽，对图像的准确率超过95%。

但是一旦我切换到视频检测，似乎连一个物体都无法检测到。

我使用以下代码进行实时目标检测：

  1. cap = cv2.VideoCapture('../video/1st/output.mp4')
  2. while cap.isOpened():
  3.     ret, frame = cap.read()
  5.     results = model(frame)
  6.     final_img = np.squeeze(results.render())
  8.     cv2.imshow("YOLO", final_img)
  9.     if cv2.waitKey(10) & 0XFF == ord("q"):
  10.         break
  11. cap.release()
  12. cv2.destroyAllWindows()

我使用以下代码加载模型：

  1. model = torch.hub.load("ultralytics/yolov5", "custom", path="yolov5/runs/train/exp36/weights/best.pt", force_reload=True)
  2. model.conf = 0.6

我甚至尝试将可用的视频拆分成JPEG图像，然后在单独的图像上运行模型，保存输出，然后将输出图像合并成一个新的视频文件。

这个方法运行得很完美，所以模型确实能够检测到一些物体。

但一旦切换到视频，似乎什么都检测不到。

英文:

I have trained a custom yolo model to detect square slots on a board, and is working with more than 95 % accuracy on images.

But as soon as I switch to video detection it seems to not detect even a single thing

I am using the following code to run real time object detection

  1. cap = cv2.VideoCapture('../video/1st/output.mp4')
  2. while cap.isOpened():
  3.     ret, frame = cap.read()
  5.     results = model(frame)
  6.     final_img = np.squeeze(results.render())
  9.     cv2.imshow("YOLO", final_img)
  10.     if cv2.waitKey(10) & 0XFF == ord("q"):
  11.         break
  12. cap.release()
  13. cv2.destroyAllWindows()

I load the model using this code

  1. model = torch.hub.load("ultralytics/yolov5", "custom", path="yolov5/runs/train/exp36/weights/best.pt",force_reload=True)
  2. model.conf = 0.6

I have even tried splitting the available video into jpegs and running the model on individual images, saving the output and then merging the output images into a new video file.

that works perfectly, so the model is detecting something.

but as soon as I switch to video it seems to go back to nothing.

答案1

得分: 0

是的，由于results.renders()返回的对象类型为none，您无法看到任何内容。
您可以像这样更改代码脚本

  1. cap = cv2.VideoCapture('../video/1st/output.mp4')
  2. while cap.isOpened():
  3.     ret, frame = cap.read()
  5.     results = model(frame)
  6.     bboxes = results.xyxy[0].cpu().tolist()
  7.     for bbox in bboxes:
  8.         conf = f'{bbox[4]:.4f}' # 预测的置信度
  9.         bbox = list(map(lambda x: int(x), bbox)) # 将浮点数转换为整数
  10.         class_id = bbox[5] # 类别ID
  11.         bbox = bbox[:4]
  12.         cv2.rectangle(frame, (bbox[0], bbox[1]), (bbox[2], bbox[3]), color=(255, 255, 255), thickness=3)
  13.     cv2.imshow("YOLO", frame)
  14.     if cv2.waitKey(10) & 0XFF == ord("q"):
  15.         break
  16. cap.release()
  17. cv2.destroyAllWindows()

并将帧写入视频。完整的代码应该如下所示

  1. input_video_path = #输入您的视频路径
  2. cap = cv2.VideoCapture(input_video_path)
  3. fps = int(cap.get(cv2.CAP_PROP_FPS))
  4. frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
  5. frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
  7. output_video_path = 'output_video.mp4' # 输出视频路径
  8. fourcc = cv2.VideoWriter_fourcc(*'XVID')
  9. out = cv2.VideoWriter(output_video_path, fourcc, fps, (frame_width, frame_height))
  11. model = #在此加载您的模型
  12. while True:
  13.     ret, frame = cap.read()
  14.     if not ret:
  15.         break
  16.     results = model(frame)
  17.     bboxes = results.xyxy[0].cpu().tolist()
  18.     for bbox in bboxes:
  19.         conf = f'{bbox[4]:.4f}' # 预测的置信度
  20.         bbox = list(map(lambda x: int(x), bbox)) # 将浮点数转换为整数
  21.         class_id = bbox[5] # 类别ID
  22.         bbox = bbox[:4]
  23.         cv2.rectangle(frame, (bbox[0], bbox[1]), (bbox[2], bbox[3]), color=(255, 255, 255), thickness=3)
  24.     out.write(frame)
  25.     if cv2.waitKey(10) & 0XFF == ord("q"):
  26.         break
  28. cap.release()
  29. out.release()
  31. # 关闭所有OpenCV窗口
  32. cv2.destroyAllWindows()

参考文献：[https://docs.ultralytics.com/][1]，[https://docs.ultralytics.com/yolov5/][2]，[https://docs.ultralytics.com/yolov5/tutorials/pytorch_hub_model_loading/][3]

英文:

Yes, You cannot see anything due to the none type of object return in the results.renders()/
You can change the code script like this

  1. cap = cv2.VideoCapture('../video/1st/output.mp4')
  2. while cap.isOpened():
  3.     ret, frame = cap.read()
  5.     results = model(frame)
  6.     bboxes = results.xyxy[0].cpu().tolist()
  7.     for bbox in bboxes:
  8.       conf = f'{bbox[4]:.4f}' #Confidance of that prediction
  9.       bbox = list(map(lambda x: int(x), bbox)) #To convert float to integer
  10.       class_id = bbox[5] #Class_id 
  11.       bbox =bbox[:4] 
  12.       cv2.rectangle(frame,(bbox[0],bbox[1]),(bbox[2],bbox[3]),color=(255,255,255),thickness=3)
  13.     cv2.imshow("YOLO", frame)
  14.     if cv2.waitKey(10) & 0XFF == ord("q"):
  15.         break
  16. cap.release()
  17. cv2.destroyAllWindows()

and write the frames in the video
the full code should look like this

  1. input_video_path = #Enter your video path
  2. cap = cv2.VideoCapture(input_video_path)
  3. fps = int(cap.get(cv2.CAP_PROP_FPS))
  4. frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
  5. frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
  7. output_video_path = 'output_video.mp4' # Output video path
  8. fourcc = cv2.VideoWriter_fourcc(*'XVID')
  9. out = cv2.VideoWriter(output_video_path, fourcc, fps, (frame_width, frame_height))
  11. model = #Here load your model
  12. while True:
  13.     ret, frame = cap.read()
  14.     if not ret:
  15.       break
  16.     results = model(frame)
  17.     bboxes = results.xyxy[0].cpu().tolist()
  18.     for bbox in bboxes:
  19.       conf = f'{bbox[4]:.4f}' #Confidance of that prediction
  20.       bbox = list(map(lambda x: int(x), bbox)) #To convert float to integer
  21.       class_id = bbox[5] #Class_id 
  22.       bbox =bbox[:4] 
  23.       cv2.rectangle(frame,(bbox[0],bbox[1]),(bbox[2],bbox[3]),color=(255,255,255),thickness=3)
  24.     out.write(frame)
  25.     #cv2_imshow(frame)
  26.     if cv2.waitKey(10) & 0XFF == ord("q"):
  27.         break
  29. cap.release()
  30. out.release()
  32. # Close all OpenCV windows
  33. cv2.destroyAllWindows()

References:-
[https://docs.ultralytics.com/][1] <br>
[https://docs.ultralytics.com/yolov5/][2] <br>
[https://docs.ultralytics.com/yolov5/tutorials/pytorch_hub_model_loading/][3]<br>

huangapple
  • 本文由 发表于 2023年8月9日 14:58:17
  • 转载请务必保留本文链接：https://go.coder-hub.com/76865316-2.html
  • conv-neural-network
  • deep-learning
  • python
  • pytorch
  • yolo
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定