I use a loop instead of data loader to read images and use model to predict them, however, my model always ouputs the same tensor. I 'm sure the input images are diffrent, and comverted to tensors with proper shape, below are the key codes, can anyone figure out the problem?
Classes = os.listdir(valdir)
for Class in Classes:
imgs = os.listdir(os.path.join(valdir, Class))
for img in imgs:
total += 1
imgPath = os.path.join(valdir, Class, img)
im = fetchImage(imgPath, trans)
im = im.cuda(0)
print(im)
output = model(im)