Skip to content

Latest commit

 

History

History
298 lines (230 loc) · 8.07 KB

README.md

File metadata and controls

298 lines (230 loc) · 8.07 KB

🖼️ VisualIndexer

AI-Powered Image Management & Semantic Search with PyTorch, CLIP, Transformers & Streamlit

Python PyTorch CLIP Transformers Streamlit TensorFlow License

📋 Project Description

VisualIndexer is a complete and intelligent system for image management, automatic indexing, and semantic search. Powered by Artificial Intelligence and state-of-the-art Deep Learning models (PyTorch, CLIP, Transformers).

This project enables:

  • 📥 Batch ingest and optimize images
  • 🔍 Automatically extract EXIF metadata
  • 📄 Recognize text in images (OCR)
  • 🏷️ Automatically generate intelligent visual tags
  • 🧠 Create semantic vector representations
  • ⚡ Search images by similarity
  • 🎨 Explore results via an interactive web interface

🚀 Key Features

1️⃣ Image Ingestion

  • Batch upload and ingest image files
  • Automatic duplicate detection (MD5 hash)
  • Intelligent optimization and resizing (max 1920x1080)
  • Adaptive JPEG compression (quality 85%)

2️⃣ Metadata Extraction

  • Complete EXIF extraction (capture date, camera, GPS, etc.)
  • Image dimensions and format
  • Automatic CSV generation for analysis

3️⃣ Text Recognition (OCR)

  • Multi-language Tesseract OCR (English + French)
  • Extract text present in images
  • JSON caching for optimization

4️⃣ Automatic Tagging

  • Vision Transformer CLIP (OpenAI)
  • Intelligent visual tag generation
  • 50+ predefined categories (city, portrait, food, document, etc.)

5️⃣ Semantic Embeddings

  • 384D vector generation with Sentence-Transformers
  • Semantic content representation
  • Advanced similarity search

6️⃣ Advanced Search Engine

  • Text search with embeddings
  • Metadata filtering (date, size, format)
  • Combined tag search
  • Intelligent result fusion

7️⃣ Interactive Web Interface

  • Modern Streamlit dashboard
  • Image visualization
  • Multi-criteria search
  • Result export

🛠️ Technologies Used

VisualIndexer uses a modern and performant technology stack:

  • Python 3.10+ - Primary language
  • Pip - Package manager

Deep Learning & Vision

Technologie Version Usage
PyTorch 2.1.1 Framework deep learning
TorchVision 0.16.1 Vision utilities
Transformers 4.35.2 HuggingFace models
Sentence-Transformers 2.2.2 Embeddings sémantiques
CLIP 0.1.0.post1 Vision-Language model

Image Processing

Technology Version Usage
Pillow 10.1.0 Image manipulation
OpenCV 4.8.1 Vision algorithms
Pytesseract 0.3.10 OCR wrapper

Data Science & Analytics

Technology Version Usage
NumPy 1.26.2 Numerical computing
Pandas 2.1.3 Dataframes & data processing
Scikit-learn 1.3.2 ML utilities

Web & UI

Technologie Version Usage
Streamlit 1.29.0 Interface web interactive

Database & Utils

Technology Version Usage
PostgreSQL - (Optional) Database
Python-dotenv 1.0.0 Environment variables
TQDM 4.66.1 Progress bars
Requests 2.31.0 HTTP client

External Infrastructure

  • Tesseract OCR - Optical character recognition (Windows/Linux/Mac)

📁 Project Structure

VisualIndexer/
├── main.py                 # Main entry point
├── requirements.txt        # Python dependencies
├── .env                    # Configuration (Tesseract path)
├── .gitignore              # Git exclusions
│
├── config/
│   └── settings.py         # Centralized configuration
│
├── scripts/                # Business logic modules
│   ├── ingest.py           # Ingestion & duplicates
│   ├── extract_metadata.py # EXIF extraction
│   ├── ocr.py              # Tesseract OCR
│   ├── tag_clip.py         # CLIP tagging
│   ├── embeddings.py       # Semantic vectors
│   └── search.py           # Search engine
│
├── ui/
│   └── interface.py        # Streamlit interface
│
├── data/
│   ├── images/
│   │   ├── raw/            # Input images
│   │   └── processed/      # Optimized images
│   ├── metadata.csv        # Metadata
│   ├── embeddings.json     # Embeddings cache
│   └── ocr_results.json    # OCR cache
│
├── models/
│   └── cache/              # ML models cache
│
├── README.md               # Documentation
├── GUIDE_UTILISATION.md    # Usage guide
└── COMMITS_GUIDE.md        # Commits guide

⚙️ Installation & Configuration

Prérequis

  • Python 3.10 ou supérieur
  • Git
  • 2GB d'espace disque (pour les modèles)

Installation Rapide

# 1. Cloner le repo
git clone https://github.com/IlyasFardaouix/VisualIndexer.git
cd VisualIndexer

# 2. Créer environnement virtuel
python -m venv venv
source venv/Scripts/activate  # Windows: venv\Scripts\activate

# 3. Installer dépendances
pip install -r requirements.txt

# 4. Installer Tesseract (Windows)
# Télécharger: https://github.com/tesseract-ocr/tesseract
# Installer et configurer path dans .env

# 5. Placer images
# Mettre images dans: data/images/raw/

# 6. Lancer le pipeline
python main.py --mode pipeline

# 7. Lancer l'interface web
python main.py --mode ui
# Accès: http://localhost:8501

Pipeline 5 Étapes

Images Brutes
    ↓
[1] INGESTION → Détection doublons, optimisation
    ↓
[2] MÉTADONNÉES → Extraction EXIF, CSV
    ↓
[3] OCR → Reconnaissance texte
    ↓
[4] TAGGING → CLIP vision, tags
    ↓
[5] EMBEDDINGS → Vecteurs sémantiques, recherche
    ↓
Résultats Indexés & Recherchables

🎯 Cas d'Usage

✅ Archivage Intelligent - Gestion massive d'images professionnelles
✅ Recherche Sémantique - Trouver images par similarité visuelle
✅ Indexation Automatique - Tags et métadonnées sans intervention
✅ Dédoublonnage - Eliminer doublons détectés
✅ Documentation - Extraire texte depuis documents scannés
✅ E-Commerce - Cataloguer produits en images

📝 Usage

Full Pipeline Mode

python main.py --mode pipeline

Processes all images in the data/images/raw/ folder

Web Interface Mode

python main.py --mode ui

Launches the Streamlit dashboard on http://localhost:8501

Ingestion Only Mode

python main.py --mode ingest

Ingests images only without AI modules

📚 Additional Documentation

💡 Optimizations & Performance

  • ✅ Intelligent ML model caching
  • ✅ Reused embedding vectors
  • ✅ Optimized JPEG compression
  • ✅ Batch processing
  • ✅ Progress tracking with TQDM

🔒 Security Configuration

Sensitive variables are stored in .env:

TESSERACT_PATH=C:\Program Files\Tesseract-OCR\tesseract.exe
OCR_LANGUAGE=eng+fra
DB_HOST=localhost
DB_PORT=5432

📄 License

MIT License - Free to use

👤 Author

Ilyas Fardaouix
GitHub: @IlyasFardaouix

🤝 Support & Contributions

Have questions or improvements? Open an Issue or submit a Pull Request

⭐ If you like this project, don't forget to star it!