About training process from scratch

[Trunk2004]

Hi @xinntao , I'm a un-graduate student, i just want to ask few thing about the repo to train a model from scratch:

1. What is "Sharpen ground-truth images during training" mean and how to replicate it?
2. Is there some factor to select the dataset for training?
3. Due to hardware issues, can i sub-sampling the image set (assume i have 2 folder hr and lq images) to feed it into the arch sequentially?
4. Same issues in 3, can the repo be clones and run on kaggle and colab? ( both have free GPU so i want to run on it if possible)
5. Finally, do you have any advice or is there something I should notice about?

That is all for now. Tks you for reading. Have a good day.

[adhirajcs]

For point 4, yes you can clone the repo and run it on kaggle and collab. In fact, in the Readme you can find a google collab link of a demo.

This link -> https://colab.research.google.com/drive/1k2Zod6kSHEvraybHl50Lys0LerhyTMCo?usp=sharing