This is a fun competition to color images.
There are two scripts
- color.py training code
- run.py run script
The python script color.py is scaffolding code to automatically load images of flower in black and white and try to learn a network to predict the original color image. We download some validation data to a directory called test, please don't modify that directory.
You are welcome to use other ML architectures besides keras, but please use wandb.log({"loss": loss}) so that we can see your performance.
At the end of the day, we will try your models on new sample images, so please make sure run.py <model_path> <input_img_path> <output_img_path> produces nice output and save your model in your wandb directory (our callback will do this autmatically).
Things to try:
- Fancier architectures
- Different loss functions
- Data augmentation
- More training data?