The opencv dnn module with cuda can't work correctly #24175
Description
System Information
OpenCV python version: 4.8.0.74
Operating System / Platform: windows 10 LTSC 21H2
Python version: 3.9.13
Cuda version: 11.7
Cudnn version: 8.4.1.50
Detailed description
I use cv2.dnn.readNetFromONNX() to load onnx file which is exported from yolov8. then set net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)
I do net.forward() and the result is [ [0,0,0,0, confidence1]
[0,0,0,0, confidence2]
[0,0,0,0, confidence3] ]
that means every bounding box's prediction is all 0 but the prediction of confidence is right. I test the python and c++ way under cuda, the problem is the same. I want the problem can be solve as soon as possible
Steps to reproduce
original_image = batch_img[i] #original_image is a element which is load from cv2.imread
[height, width, _] = original_image.shape
length = max((height, width))
image = np.zeros((length, length, 3), np.uint8)
image[0:height, 0:width] = original_image
scale = length / 640
blob = cv2.dnn.blobFromImage(image, scalefactor=1 / 255, size=(640, 640), swapRB=True)
#cv2.cuda.setDevice(0)
net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)
net.setInput(blob)
outputs = net.forward()
Issue submission checklist
- I report the issue, it's not a question
- I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
- I updated to the latest OpenCV version and the issue is still there
- There is reproducer code and related data files (videos, images, onnx, etc)