Hi everyone,

First, congratulations to all winners and participants of this challenge.

I have not taken part in this competition due to time constraints but I have been working recently on developing a tool for data augmentation in Keras that allows the use of hyperspectral images. As we all know, the current implementation of ImageGenerator Keras only allows up to 4 channels which is not ideal and that is due to the fact that they use Pillow library for image processing.

I have developed this open tool for augmentation that uses sklearn for image processing and allows numerous channels for image processing. It is work in progress as I plan to make it more suitable for satellite tiled images in the future and allow random crops from the original tiled images but in the meantime, feel free to experiment with it and let me know if you face any difficulties.

https://github.com/mabdelhack/hyperspectral_image_generator