convert cv2.umat to numpy array

Question

Processed_image() function returns a cv2.Umat type value which is to be reshaped from 3 dimensions(h, ch, w) to 4 dimensions(h, ch, w, 1) so i need it to be converted to numpy array or also if possible help me to directally rehshape cv2.umat type variable to be directally reshaped and converted to a pytorch tensor and can be assigned to reshaped_image_tensor.

img_w=640
img_h=640
img_ch=3
umat_img = cv2.UMat(img)
display_one(umat_img, "RESPONSE")    #function created by me to display image
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
with torch.no_grad():
    processed_img = preprocess_image(umat_img, model_image_size = (img_h, img_ch, img_w))
    #___________write YOUR CODE here________
    reshaped_images_tensor = torch.from_numpy(processed_img.reshape(img_h, img_ch, img_w, 1)).float().to(device)      #images_tensor.reshape(img_h, img_ch, img_w, 1)
    outputs = model(reshaped_images_tensor)
    _, predicted = torch.max(outputs, 1)
    c = predicted.squeeze()
    output_probability(predicted, processed_img, umat_img)
if ord('q')==cv2.waitKey(10):
    exit(0)

Answer 1

I didn't quite catch your question, but you can get numpy data of an opencv's umat with "get()" like this

and you should probably permute your tensor before feeding it into your model.