Fix: Implement path portability and add default resolution to train.py and data.py

[Shrishagk]

Overview

This PR improves the robustness of the training pipeline by addressing path resolution issues and preventing null-pointer errors during the PSF kernel setup in Grid_based_strong_lensing_for_unsupervised_super_resolution_Anirudh_Shankar/data.py and Grid_based_strong_lensing_for_unsupervised_super_resolution_Anirudh_Shankar/train.py

Changes

1.Path Portability:

Introduced script_dir using os.path.dirname(os.path.abspath(file)).

Defined grid_dir to explicitly point to the grid_matrices folder.

Updated all torch.load calls to use os.path.join(), ensuring the script can find its dependencies regardless of the working directory.

2.Argparse Improvements:

Added default=0.1 to the --resolution argument. This ensures that the physical lensing calculations have a valid numerical input even if the user does not manually specify the resolution.

3.Updated data.py:

To use os.path.join for dataset indexing. This ensures robust path construction regardless of whether the input directory string contains trailing slashes and improves cross-platform compatibility.

Impact

These changes make the repository much easier to use for new contributors, as the "out-of-the-box" execution of train.py no longer requires manual file moving or specific terminal positioning.

Testing

Verified by running: python Grid_based_strong_lensing_for_unsupervised_super_resolution_Anirudh_Shankar/train.py from the repository root. The script now correctly identifies the matrices and initializes the PSF kernel without errors.