U-Net Paper Workthrough

Author(s): Fatma Elik

Originally published on Towards AI.

Convolutional Networks for Biomedical Image Segmentation Document Explanation in Detail

This member-only story is on us. Upgrade to access all of Medium.

Convolutional networks have been around for a long time, but their performance has been limited by the size of the available training sets and the size of the networks under consideration.

The study by Krizhevsky et al. was a great achievement. They trained a huge network with 8 layers and millions of parameters using 1 million training images on the ImageNet dataset. Since then, even larger and deeper networks have been trained.

Convolutional networks are mostly used for classification problems. Here the output is a single class label out of an image. However, for computer vision challenges especially for biomedical image processing, the intended output should contain localization, which means assigning a class label to each pixel.

Sliding window approaches aid in estimating the class label of each pixel by using a local region (patch) around that pixel as input. The network won the EM segmentation challenge at ISBI 2012 due to its localization capabilities and larger training data in patches than images.

The network is slower due to redundancies and trade-offs between localization accuracy and context usage. Newer techniques propose a classification output considering many layers, allowing… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI