FaceBoxes.PyTorch
FaceBoxes.PyTorch copied to clipboard
reduce numbers of prior_box
I am currently trying to reduce the number of prior_boxes. The current approach is to reduce the dense anchor ratio to 2:1:1 (originally 4:2:1). This scheme is currently available on my data set and ap improved (about 1 point), but the precision dropped more (3-4 points), is there any way to keep the precision not significantly reduced?