U-Net Architecture

NON EXAMINABLE!

U-Net is an architecture for image semantic segmentation, splitting an image into meaningful parts. The architecture is split into two parts:

• Contracting Path (Encoder) - captures context using convolutional layers and downsampling (max pooling).
• Expanding Path (Decoder) - enables precise localization using upsampling and skip connections.

Skip Connection concatenates the feature maps from the contracting path to the corresponding layers in the expanding path, preserving spatial information lost during downsampling.