LumiNet: Single-Stage CNN for Low-Light Enhancement and Denoising

LumiNet is a compact single-stage CNN developed for joint low-light image enhancement and denoising. Poor illumination, image noise, colour distortion, and detail loss are addressed within a single forward pass, enabling efficient image restoration while avoiding the complexity often found in multi-stage approaches. The project focuses on lightweight design, practical deployment, and real-world usability for efficient image restoration in low-light conditions.

• 🌐 Live UI
• 🤗 Hugging Face Space

High-Level Architecture

Model Profile

Saved evaluation run:

• Parameters: 1.49M
• FLOPs: 6.83 GFLOPs at 256x256

Test-Split Evaluation

The table below shows saved evaluation results.

Dataset	Task	PSNR	SSIM	LPIPS	Exposure Error	Colour Error
DLL	Enhancement	18.88	0.5714	0.3307	0.0873	0.0120
LOLv2-Real	Enhancement	21.13	0.8422	0.1783	0.0713	0.0138
LSRW	Enhancement	17.67	0.5631	0.2657	0.1013	0.0138
SIDD	Denoising	35.23	0.9123	0.1760	-	0.0048
NLL	Joint	17.97	0.5598	0.3667	0.0988	0.0126

Repository Layout

• backend/: training, evaluation, notebooks, outputs, and Python tests
• configs/: dataset, model, training, evaluation, and ablation configs
• integrations/: Hugging Face Space bundle and copied runtime model files
• ui/: Next.js frontend and API routes
• app.py: root launcher that reuses integrations/app.py

Requirements

• Python 3.12 or newer
• Node.js 18 or newer
• npm

How to Use

1. Run the UI

Step 1: Clone the repository

git clone https://github.com/Himidiri/LumiNet.git

Step 2: Go to the UI folder

cd LumiNet/ui

Step 3: Install the UI packages

npm install

Step 4: Start the UI

npm run dev

Open http://localhost:3000.

2. Run the backend locally

Use these steps when local LumiNet inference is needed for the web UI or for direct local testing.

Step 1: Go to the project root

cd LumiNet

Step 2: Create a Python virtual environment (recommended)

python3 -m venv .venv

Step 3: Activate the virtual environment

source .venv/bin/activate

Step 4: Install the backend packages

pip install -r integrations/requirements.txt

Step 5: Start the local Gradio app

python app.py

python3 -m venv .venv helps keep LumiNet packages separate from other Python projects.

source .venv/bin/activate is needed only when .venv is used. If the required packages are already installed in another Python environment, Step 2 and Step 3 can be skipped.