yolov5-face
yolov5-face copied to clipboard
Zero precision Recall
Hi, I've recently changed your code to find more landmarks on face. but during training the P, R, map are zero. I figure out the output of this line:
inf_out, train_out = model(img, augment=augment)
return an inf_out variable that it's class probability is 0!
this issue cause problem in non_max_suppression_face() function and the force all conf scores to zero.
so eventually there is no detection result.
what is the difference between these two line?
test.py ==== >inf_out, train_out = model(img, augment=augment)
train.py =======> with amp.autocast(enabled=cuda): pred = model(imgs)