Semantic Segmentation Training

1x1 kernel won’t be good. You should choose a kernel size divisible by stride to avoid artifacts.

1 Like