Closed JadBatmobile closed 5 years ago
Due to the padding operations done in the earlier layers of FCN network (see, for example, https://github.com/sadeepj/crfasrnn_keras/blob/master/src/crfrnn_model.py#L46), the output segmentation masks/logits have a boarder around them and are not aligned with the input image. Cropping2D layers are used to remove these boarders. Please refer to the FCN paper for more details.
Can you explain the purpose of the Cropping2D layers?