This repository consists Inception Unet versions of classical Unet architecture for image segmentation. In the paper, a new deep learning architecture has been developed by combining inception blocks with the convolutional layers of the original U-Net architecture to achieve remarkably high performance in building detection.
You can train your model by using [Massachusetts Buildings Dataset] https://www.cs.toronto.edu/~vmnih/data/
To train Unet, Inception or UnetV2 model
import unet, Inception, unetV2
x, y = ... # range [0,1] normalized images and ground truth map
model = unetV2.get_unet_plus_inception()
model.compile(optimizer=Adam(lr=1e-5), loss=dice_coef_loss, metrics=[dice_coef])
model.fit(x,y)
If you use this work in your publications, please cite it as below:
@article{delibasoglu2020improved,
title={Improved U-Nets with inception blocks for building detection},
author={Delibasoglu, Ibrahim and Cetin, Mufit},
journal={Journal of Applied Remote Sensing},
volume={14},
number={4},
pages={044512},
year={2020},
publisher={International Society for Optics and Photonics}
}