keras-unet-collection 0.0.5

The tensorflow.keras implementations of U-net, U-net++, R2U-net, Attention U-net, ResUnet-a ,and U^2-Net.

keras-unet-collection

This repository contains tensorflow.keras implementations of U-net, U-net++, R2U-net, Attention U-net, ResUnet-a, U^2-Net:

keras_unet_collection.models	Name	Reference
unet_2d	U-net/Unet	Ronneberger et al. (2015)
unet_plus_2d	U-net++/Unet++	Zhou et al. (2018)
r2_unet_2d	R2U-Net	Alom et al. (2018)
att_unet_2d	Attention U-net	Oktay et al. (2018)
resunet_a_2d	ResUnet-a	Diakogiannis et al. (2020)
u2net_2d	U^-Net	Qin et al. (2020)

These models are implemented with user-friendly hyper-parameter options and keywords, including network depth, hidden layer activations and batch normalization. User guide provided several examples.

Additional activation layers and loss functions are also provided:

keras_unet_collection.activations	Name	Reference
GELU	Gaussian Error Linear Units (GELU)	Hendrycks et al. (2016)
Snake	Snake activation	Liu et al. (2020)

keras_unet_collection.losses	Name	Reference
tversky	Tversky loss	Hashemi et al. (2018)
focal_tversky	Focal Tversky loss	Abraham et al. (2019)
crps2d_tf	CRPS loss (experimental)

Dependencies

• TensorFlow 2.3.0

• Keras 2.4.0

• Numpy 1.18.2

Installation and usage

pip install keras-unet-collection

from keras_unet_collection import models
# e.g. models.unet_2d(...)

• Jupyter notebooks are provided as user guides.

Note: Because of the flexible hyper-parameter options, neural networks produced by this package may not be compatible with other pre-trained models of the same name/kind. Training from scratch is recommended.

• Changelog

Overview

U-net is a convolutional neural network with encoder-decoder architecture and skip-connections, loosely defined under the concept of "fully convolutional networks." U-net was originally proposed for the semantic segmentation of medical images and is modified for solving a wider range of gridded learning problems.

U-net and many of its variants take three or four-dimensional tensors as inputs and produce outputs of the same shape. One technical highlight of these models is the skip-connections from downsampling to upsampling layers, which benefit the reconstruction of high-resolution, gridded outputs.

Contact

Yingkai (Kyle) Sha [e-mail]

License

MIT License