Method: apply transfer learning because of limited training time on Colab

[JohnTigue]

Something like imagenet, or maybe U-Net has pre-trained models. Basically, transfer learning would greatly reduce the training time, making training on Colab a realistically pleasant experience.

[JohnTigue]

Note that any color pretrained model will need to address the grayscale nature of the data. See

• Section 2.5: Training with non-RGB data

[JohnTigue]

Transfer Learning in Keras with Computer Vision Models (2019)

[JohnTigue]

Convolutional Neural Networks for Multi-class Histopathology Image Classification, 2019, arxiv pre-print, ResNet transfer learning, PyTorch