sawabedarain/Lumina-IQA
🖼️ Image Quality Assessment – Deep Learning Pipeline
<p align="center"> <img src="assets\banner.jpeg" alt="Image Quality Assessment Banner" width="800"/> </p>
<p align="center"> <a href="https://fastapi.tiangolo.com/"><img src="https://img.shields.io/badge/API-FastAPI-009688?logo=fastapi" /></a> <a href="https://pytorch.org/"><img src="https://img.shields.io/badge/Framework-PyTorch-EE4C2C?logo=pytorch" /></a> <a href="https://www.docker.com/"><img src="https://img.shields.io/badge/Container-Docker-2496ED?logo=docker" /></a> <a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/License-MIT-blue.svg" /></a> </p>
A full-stack project for No-Reference Image Quality Assessment (NR-IQA). Built with Python, PyTorch, FastAPI & Docker, this pipeline takes raw images, runs a trained deep-learning model, and predicts a quality score without needing the original reference image.
📜 Table of Contents
🔎 About
Real-world images often suffer from compression, blur, noise, or artifacts. Traditional IQA metrics (PSNR, SSIM) need a “perfect” reference image — which we rarely have. This project implements deep learning NR-IQA to:
- Predict perceptual quality from any image
- Enable automated QA in photo platforms, streaming, and dataset filtering
Dataset used: [KonIQ-10k](http://database.mmsp-kn.de/koniq-10k-database.html) (10,000 diverse images, 512×384 px)
✨ Features
- 🚀 FastAPI REST endpoint for real-time scoring
- 🧠 PyTorch model with configurable backbone
- 🎨 Modern UI for uploading images & visualizing scores
- 📦 Docker-ready for clean deployment
- ☁️ Render-friendly: automatic builds & deploys
- 📊 Hooks for model retraining & logging
🖥️ UI Preview
<p align="center"> <img src="static\ui_preview.png" alt="UI Preview" width="700"/> </p>
The /ui endpoint serves a clean, user-friendly web interface for image uploads & quality scoring.
📂 Project Structure
ImageQualityAssessment/ │ ├── artifacts/ # Saved models (model.pth etc.) ├── data/ # (ignored by git) dataset / images ├── src/ │ ├── model.py # PyTorch model & loading utilities │ ├── inference_api.py # FastAPI app exposing /predict + UI │ └── utils.py # Preprocessing & helpers │ ├── static/ # HTML, CSS, JS for UI │ └── index.html │ ├── requirements.txt # Python dependencies ├── Dockerfile # Build instructions for Docker ├── .dockerignore # Ignore unnecessary files for container ├── .gitignore # Ignore venv, data, logs etc. ├── README.md # You’re reading it └── LICENSE
⚙️ Installation
1. Clone the repo
<pre> ``bash git clone https://github.com/DarainHyder/Image_Quality_Assessment.git` </pre> <pre> `bash cd Image_Quality_Assessment `` </pre>
2. Create & activate virtualenv (optional but clean)
<pre> ``bash python -m venv venv ` </pre> <pre> `bash source venv/bin/activate # (Windows: venv\Scripts\activate) `` </pre>
3. Install dependencies
<pre> ``bash pip install -r requirements.txt ` </pre> 🚀 Usage Run API locally <pre> `bash uvicorn src.inference_api:app --host 0.0.0.0 --port 8000 --reload `` </pre>
API Docs → http://127.0.0.1:8000/docs
UI Page → http://127.0.0.1:8000/ui
🐳 Docker Deployment
Build Docker image
docker build -t image-quality-assessment .
Run container
<pre> ``bash docker run -d -p 8000:8000 image-quality-assessment `` </pre> Visit:
API Docs → http://localhost:8000/docs
UI Page → http://localhost:8000/ui
🛠 Tech Stack
Python 3.11
PyTorch – Deep learning framework
FastAPI – High-performance web API
Uvicorn – ASGI server
Docker – Containerization & cloud deployment
HTML/CSS – Custom UI frontend
📜 License
MIT – feel free to fork & remix. If you use this repo, star ⭐ it and tag me — love seeing real-world uses!
Clean code, reproducible pipelines, and a Docker-first mindset — ship models like products. 🚀