signifance

Hello author, I would like to ask how you can determine the gradient size of each pixel in the model for an input original image

nips_gd.py This code：print('[{it:.2f}][{name}] ' 'BTOP1: {BTOP1.avg:.2f} ' 'lossX: {ori:d}/{redu:d} ' 'Time: {ti:.3f} ' 'L1: {l1:.1f} ' 'M&m {mean:.2f}/{median:.2f} ' 'T&t {tmean:.2f}/{tmedian:.2f} ' 'Num: {num}'.format( it = float(idx*100)/len(test_loader), name = image_names[0].split('_')[-1],...