SAGA-SR

This is the implementation of SAGA-SR.

⚠️ This code is under construction. The repository currently contains the preprocessing and training code only, and may still contain errors.

TODO

• Update requirements.txt
• Upload pretrained model weights
• Add inference / evaluation code

Installation

This project is built on top of TorchJaekwon, which must be installed first.

# Clone TorchJaekwon
git clone https://github.com/jakeoneijk/TorchJaekwon
cd TorchJaekwon

# Install TorchJaekwon
source install_torchjk.sh

Usage

The entry point follows the TorchJaekwon convention. Pass the config file with --config_path and the stage with --stage:

python main.py --config_path CONFIG_PATH --stage STAGE_NAME
# STAGE choices: ['preprocess', 'train', 'inference', 'evaluate']

Config files live under config/. The reference config is config/saga_sr.yaml.

Preprocess

python main.py --config_path config/saga_sr.yaml --stage preprocess

Train

python main.py --config_path config/saga_sr.yaml --stage train