lwbna-unet 1.0.2

TF2 (Keras) implementation of LWBNA_Unet. Unrelated to the authors of the paper

Light Weight Bottle Neck Attention Unet

TF implementation of the architecture described in A lightweight deep learning model for automatic segmentation and analysis of ophthalmic images by Sharma et al.

This is an independent implementation unrelated to the autors of the paper. I have used it for segmenting fibers in my own project. Please leave a Star if this code is useful to you :smile:.

Usage

# install your favorite version of tensorflow2
pip install tensorflow
# install this package
pip install lwbna-unet

import lwbna_unet as unet
import numpy as np

# input has shape `(Batch size, Height, Width, Channels)`
# input has dtype float and is expected to be normalized to the range [0,1].
# output has shape `(Batch size, Height, Width, n_classes)`

my_unet = unet.LWBNAUnet(
    n_classes=1, 
    filters=128, 
    depth=4, 
    midblock_steps=4, 
    dropout_rate=0.3, 
    name="my_unet"
)

# the network is untrained. Dummy input.
my_unet.build(input_shape=(8,320,320,3))
my_unet.predict(np.random.rand(8,256,256,3))
my_unet.summary()
# you can now train `my_unet` as a regular `keras.Model`

<script async defer src="https://buttons.github.io/buttons.js"></script>