1. The overall structure of UNet
Simply put, it is the process of encoding and decoding. And also introduces the feature splicing operation.
Unet consists of two parts.
The first part is feature extraction, which is similar to VGG. Every time the feature extraction part passes through a pooling layer, it is a scale, including a total of 5 scales including the original image scale.
The second part is the upsampling part. The upsampling part is spliced at the same scale as the number of channels corresponding to the feature extraction part every time it is sampled, but it must be cropped before splicing.
2. The role of copy and crop
The shallow network will save obvious content information. The deeper the network layer, the less content and more features. In order to add content information to the deep network, copy and crop operations are performed.