👆 DirectIR: Training-Free Enhancement for Diffusion-Based Image Restoration via Contrastive Guidance

Image Processing 2024 Final Project

Group 5: Group-SEVEN

Youtube DEMO LINK

Environment Setup

Follow the steps below to clone the repository, set up the environment, and install dependencies. The code is tested on Python 3.10.

git clone https://github.com/yeeecheng/directIR.git
cd directIR
conda create -n directIR python=3.10 -y
conda activate directIR
pip install -r requirements.txt

Inference

To process the default image examples, run the following command. The pre-trained model weights will be automatically downloaded from Hugging Face and placed under the checkpoints/ directory:

bash scripts/inference.sh

Overall Pipeline

Overview

Textual Contrastive Loss

Experimental Results

	Degraded Image	Ground truth	PromptFix	PromptFix w/ Ours
Low-Light
Motion-Blur
Snow

🙏 Acknowledgments

We would like to thank the authors of Promptfix, InstructDiffusion, Stable Diffusion, and InstructPix2Pix for sharing their codes.