Clarification on SIDD Dataset Setup and Usage in NAFNet

[hij1112]

Hi, I've been reviewing the guidelines for setting up the SIDD dataset and have some questions:

1. Use of Official Dataset for Training and Validation in NAFNet:

   • It seems like all images from the official dataset are being used both for training and validation. Specifically, the training process uses cropped images from all original images, with a subset selected for validation. Is this understanding correct? If so, could you clarify why the dataset wasn't strictly divided for training and validation purposes, such as by scene number, smartphone code, or another criterion?

2. Selection of Validation Dataset:

   • The provided validation dataset appears to include only 8 selected scenes out of the total 10 available scenes. What was the rationale behind selecting these specific scenes for validation? Can you provide a code on how you produced the validation dataset?

3. Different Image size of Training/Validation Dataset:

   • The provided validation dataset comprises 256x256 images, while the data preparation code generates 512x512 images for the training dataset. What are the reasons for this setup?

Any clarification on these points would be greatly appreciated. Thank you for your time and assistance.