InpaintReward

October 10, 2025 · View on GitHub

This project implements a reward model designed for image inpainting tasks, aiming to evaluate the quality of inpainted results based on learned representations.

Dataset Preparation

  1. Download the training dataset from this Google Drive link.
  2. Place the dataset contents in the ./Data/ directory.

🚀 Usage

cd ImageReward
cd train
In the YAML file located at `src/config/config.yaml`, replace: ../Data/ with your own path
Run bash ./scripts/train_one_node.sh
You can monitor training progress by checking the log files in director "./log"