VIIS: Visible and Infrared Information Synthesis for Severe Low-light Image Enhancement

This repository contains the official implementation of the paper
"VIIS: Visible and Infrared Information Synthesis for Severe Low-light Image Enhancement".

---

🚀 Getting Started

Follow the steps below to set up the environment and prepare all necessary resources.

1. Install the Environment

Set up the environment according to the instructions provided in the
CompVis latent-diffusion repository.

git clone https://github.com/CompVis/latent-diffusion.git
cd latent-diffusion
conda env create -f environment.yaml
conda activate ldm

Return to the VIIS directory after the environment is set up.

---

2. Install Deformable Attention Module

Install the attention mechanism discrepancy from
Deformable-DETR:

git clone https://github.com/fundamentalvision/Deformable-DETR.git
cd ./Deformable-DETR/models/ops
sh ./make.sh

---

3. Download Checkpoint

Download the pre-trained VIIS model checkpoint on msrs dataset from Google Drive
and place it in the ./checkpoint/ directory.

---

4. Download VQ-F4 First-Stage Model

Download the VQ-F4 model from the latent-diffusion releases
and place it in the ./models/first_stage_models/ folder.

---

5. Download MSRS Dataset

Download the MSRS dataset from the PIAFusion repository
and place it in the ./data/ directory.