- create dataloader
- design transformations that are consistent to train/test images and scale them down/crop to small, consistent image.
- create transformations (do not destroy the pictures!)
- linear
- exponential
- logarithmic
- gamma adjustment (opencv)
- linear mapping function (https://www.cs.cityu.edu.hk/~rynson/papers/eccv22b.pdf)
- ask Morten for tips on aproaching weight initialization in stable diffusion
- implement the stable diffuction model (ground up? using some weights?)
- pick and create metrics
- try training
- train using different transformations
- measure the results and add them to table
- compare with some papers
- compare visually
- make a poster