Low Light Rainy Code Export

This folder is a portable code-only export of Low_light_rainy_new.

Excluded

The following large/runtime folders were intentionally excluded:

dataset/
checkpoint*/
results/
Python caches
.pth, .zip, .log files

Included Core Code

Key files for the current restoration experiments:

train_restormer.py - main trainer for baseline, v1, and v2.
restormer.py - original Restormer backbone and existing variants.
net/restormer_lowlight_rain.py - FGP-Restormer v1.
net/restormer_lowlight_rain_v2.py - FGP-Restormer v2.
utils/inference_utils.py - validation padding, crop, RGB PSNR, optional TTA.
run_restormer_fgp.sh - v1 launch script.
run_restormer_fgp_v2.sh - v2 launch script.
dataset_RGB_ori.py, data_RGB.py - legacy data loaders.
dataset_rwjr10k.py, train_restormer_rwjr10k.py, train_restormer_rwjr10k_from_original.py - RWJR-10K related code, if needed.

Environment

The current new conda environment was exported to:

requirements_new.txt
environment_new.yml

Recommended setup on a new server:

conda env create -f environment_new.yml
conda activate new

If the environment name conflicts, edit the first line of environment_new.yml, or use pip inside your own environment:

python -m pip install -r requirements_new.txt

Data Layout Expected by Current v1/v2 Scripts

Default low-light-rain training expects:

dataset/train/syn+real/input
dataset/train/syn+real/target
dataset/test/input
dataset/test/target

The v1/v2 scripts do not include dataset files in this export. Copy datasets separately to the same relative locations or pass explicit paths:

python train_restormer.py \
  --train_inp /path/to/input \
  --train_tar /path/to/target \
  --test_inp /path/to/test/input \
  --test_tar /path/to/test/target

Run v1

bash run_restormer_fgp.sh

Run v2

bash run_restormer_fgp_v2.sh

Before running on a new server, update any --resume paths in the run scripts if checkpoints are not present.