Hi team,
I have applied unfold function to patchify res3 feature maps with overlaps (kernel=3).then I fold it back. The accuracy of deep learning model has increased.i couldn’t figure out why the accuracy has increased.does fold function with overlaps increase the receptive field ?