README.md

metadata

title: Image Search with ColModernVBert
emoji: 🔍
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false

Image Search with ColModernVBert

A multimodal image search demo using ColModernVBert for cross-modal retrieval between text queries and images.

Features

• Multimodal Search: Query images using natural language text
• ImageNet-1K Dataset: Searches through 1000 diverse validation images
• Real-time Indexing: Automatically indexes 1000 images on startup
• 512x512 Optimization: Images resized for optimal model performance
• Multiprocessing: Fast parallel image preprocessing
• Upload Support: Upload custom images or switch datasets

Usage

Enter text queries like:

• "dog" → Various dog breeds
• "sports car" → Different car models
• "musical instrument" → Guitars, pianos, violins
• "food" → Fruits, vegetables, dishes
• "nature" → Trees, flowers, landscapes

Adjust the Top-K slider to control the number of results returned.

Technical Details

• Model: ColModernVBert (ModernVBERT/colmodernvbert)
• Dataset: ImageNet-1K validation set (1000 images)
• Image Size: 512x512 pixels
• Embeddings: Cosine similarity between text and image embeddings

Performance

• Indexing: ~2-3 minutes for 1000 images
• Search: Near-instant results
• Memory: Optimized for Space hardware limits

---

Built with Gradio and ColModernVBert for demonstration purposes.