Follow-up to #135: Add dataset validation checks and config file for reproducibility #196

Follow-up to #135

PR #136 resolved the core dataset setup issues. During the discussion on #135,
I suggested two additional improvements that the maintainer agreed would further
improve robustness and reproducibility.

Proposed Improvements

1. Dataset Validation Script (validate_dataset.py)
   Before training begins, a lightweight validation step should verify:

• HR and LR image counts match
• Filenames are consistent across HR and LR folders
• Image shapes are as expected (based on scale factor)

This prevents subtle misalignment errors that won't raise immediate exceptions
but silently corrupt training.

2. Configuration File (config.yaml)
   Dataset-related parameters like scale factor and patch size are currently
   implicit/hardcoded. Externalizing them into a config file would:

• Make experiments more transparent
• Improve reproducibility across different setups
• Allow contributors to change parameters without touching core training logic

Reference

• Closes follow-up from Dataset setup is unclear and causes errors when running Super Resolution pipeline #135
• Maintainer acknowledged these suggestions in the issue thread

I'd like to work on this
I'm a GSoC 2026 applicant and happy to submit a PR implementing both of these.
Can I be assigned?