The encoder downsamples the input image to grasp its intricate details, while the decoder upsamples to reconstruct ... layers provide global attention, CNN filters contribute to local attention.