Since the paper mentions that for non-rigid edits we should choose “the layers that rely more on content similarity”, we would expect to select the blocks with higher PSNR values as shown in Figure 2. However, I found that the selected blocks [0, 7, 8, 9, 10, 18, 25, 28, 37, 42, 45, 50, 56] are not all those with high PSNR — some of them even have relatively low PSNR values. Could you please explain why?
Thank you.
Since the paper mentions that for non-rigid edits we should choose “the layers that rely more on content similarity”, we would expect to select the blocks with higher PSNR values as shown in Figure 2. However, I found that the selected blocks [0, 7, 8, 9, 10, 18, 25, 28, 37, 42, 45, 50, 56] are not all those with high PSNR — some of them even have relatively low PSNR values. Could you please explain why?
Thank you.