U-Net

Example_architecture_of_U-Net_for_producing_k_256-by-256_image_masks_for_a_256-by-256_RGB_image

U-Net is a type of convolutional neural network (CNN) primarily used for image segmentation. It was first introduced by Olaf Ronneberger, Philipp Fischer, and Thomas Brox in 2015 for the task of biomedical image segmentation. The architecture of U-Net is designed to work with very few training images and to yield more precise segmentations.

Architecture

The U-Net architecture consists of a contracting path (encoder) and an expansive path (decoder), which gives it the U-shape. The encoder is a typical CNN that consists of repeated application of two 3x3 convolution layers, each followed by a rectified linear unit (ReLU) and a 2x2 max pooling operation for downsampling. The decoder path consists of upsampling of the feature map followed by a 2x2 convolution that halves the number of feature channels, a concatenation with the corresponding cropped feature map from the encoder path, and two 3x3 convolutions, each followed by a ReLU.

Contracting Path

The contracting path follows the typical architecture of a convolutional neural network. It consists of the repeated application of two 3x3 convolutions (unpadded convolutions), each followed by a ReLU and a 2x2 max pooling operation with stride 2 for downsampling. At each downsampling step, the number of feature channels is doubled.

Bottleneck

The bottleneck is the layer at the bottom of the U, where the feature maps are the smallest in spatial dimensions but have the highest number of channels. This layer connects the contracting path to the expansive path.

Expansive Path

The expansive path consists of an upsampling of the feature map followed by a 2x2 convolution ("up-convolution") that halves the number of feature channels, a concatenation with the correspondingly cropped feature map from the contracting path, and two 3x3 convolutions, each followed by a ReLU. The cropping is necessary due to the loss of border pixels in every convolution.

Applications

U-Net has been widely used in various fields, particularly in biomedical image segmentation, where it has achieved state-of-the-art performance. It is also used in other areas such as satellite image analysis, autonomous driving, and medical imaging.

Advantages

• **Data Efficiency**: U-Net can work with a limited amount of training data.
• **Precision**: It provides precise segmentation maps.
• **Flexibility**: It can be applied to various types of image segmentation tasks.

See Also

• Convolutional neural network
• Image segmentation
• Biomedical image analysis
• Deep learning
• Machine learning