U-Net is a convolutional architecture for semantic segmentation introduced by Olaf Ronnerberg et al. in Convolutional Networks for Biomedical Image Segmentation with the explicit goal of segmenting biomedical images.
The architecture revealed itself to be general enough to be applied in every semantic segmentation task since it has been designed without any constraints about the datatypes.
The U-Net architecture follows the typical encoder-decoder architectural pattern with skip connections. This way of designing the architecture has proven to be very effective when the goal is to produce an output with the same spatial resolution of the input since it allows the gradients to propagate between the output and the input ...
