{"id":9544,"date":"2026-07-27T12:30:28","date_gmt":"2026-07-27T10:30:28","guid":{"rendered":"https:\/\/www.kubicek.ai\/?post_type=lexicon&#038;p=9544"},"modified":"2026-07-27T13:23:26","modified_gmt":"2026-07-27T11:23:26","slug":"u-net","status":"publish","type":"lexicon","link":"https:\/\/www.kubicek.ai\/en\/lexicon\/u-net\/","title":{"rendered":"U-Net"},"content":{"rendered":"<p class=\"wp-block-paragraph\"><strong>U-Net<\/strong> is a convolutional architecture designed in 2015 for the segmentation of biomedical images, which became the reference solution for that whole task. Its name comes from the shape of its diagram: the descending left branch (the contracting path) progressively downsamples the resolution and increases the number of channels, thereby gaining an understanding of context, while the ascending right branch symmetrically restores the resolution back to the original size. The key innovation is the skip connections, which link corresponding levels of the two branches and carry fine spatial detail into the decoder that would otherwise be lost irretrievably during downsampling; without them the resulting masks would be blurred and imprecise in their outlines. The architecture is moreover fully convolutional, so it has no fixed input size, and it was conceived from the outset for work with very small datasets \u2013 which is exactly why it relies on strong data augmentation, especially elastic deformations. Besides medicine it is used in satellite imaging, microscopy and industrial inspection, and in modified form it constitutes the denoising core of most diffusion models for image generation.<\/p>\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n<p class=\"wp-block-paragraph\">Imagine a restorer who has to mark the damaged spots on a large fresco. First he walks far away from the wall to see the whole composition and understand what the fresco depicts \u2013 but in doing so he loses the detail. Then he slowly comes back closer and fills in the fine points. If on the way back he did not remember what he saw from each distance, he would end up with blurry blobs. So at every step back he takes a photograph, and reviews them again on the way in. Those photos from the outward journey, used on the return, are precisely the skip connections that make U-Net work.<\/p>\n","protected":false},"featured_media":0,"template":"","class_list":["post-9544","lexicon","type-lexicon","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon\/9544","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon"}],"about":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/types\/lexicon"}],"wp:attachment":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/media?parent=9544"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}