---
title: Image Search
sdk: gradio
emoji: 📚
colorFrom: pink
colorTo: red
---


# Image Search with SigLIP2

An image search application powered by Google's SigLIP2 model (`google/siglip2-so400m-patch16-naflex`).

## Features

- 🔍 Natural language image search
- 🖼️ Search through a curated database of images
- 📊 Similarity scores for each result
- 🎯 Adjustable number of results (top-k)
- 📁 Easy image management via Excel spreadsheet

## How It Works

The app uses the SigLIP2 vision-language model to:
1. Load image URLs from an Excel spreadsheet (`image_database.xlsx`)
2. Encode all images in the database into embeddings
3. Encode your text query into an embedding
4. Find images with the highest similarity to your query
5. Display the top matching results with similarity scores

## Image Database Format

The app reads image URLs from an Excel file named `image_database.xlsx`. The Excel file should have:

- **Required:** A column named `url` (or `URL`, `image_url`, `urls`, `link`, or `image`) containing the image URLs
- **Optional:** Additional columns like `description`, `category`, etc. for your own reference

### Example Excel Format:

| url | description |
|-----|-------------|
| https://example.com/image1.jpg | Mountain landscape |
| https://example.com/image2.jpg | Cat photo |
| https://example.com/image3.jpg | Beach sunset |

### To Update Your Image Database:

1. Edit `image_database.xlsx` with your own image URLs
2. Save the file
3. Restart the Gradio app

The app will automatically load all URLs from the Excel file at startup.

## Usage

1. Enter a search query (e.g., "sunset", "cat", "mountains", "technology")
2. Adjust the number of results you want to see
3. Click "Search" or press Enter
4. View the matching images with their similarity scores

## Example Queries

- "a cat"
- "mountain landscape"  
- "beach sunset"
- "city skyline at night"
- "person running"
- "food on a plate"
- "car on the road"
- "flowers in bloom"

## Model

This space uses the **google/siglip2-so400m-patch16-naflex** model, a state-of-the-art vision-language model from Google.

## Local Setup

To run this locally:

```bash
pip install -r requirements.txt
python app.py
```

Make sure you have `image_database.xlsx` in the same directory.

## Deployment on Hugging Face Spaces

1. Create a new Space on Hugging Face
2. Select "Gradio" as the SDK
3. Upload `app.py`, `requirements.txt`, and `image_database.xlsx`
4. The Space will automatically build and deploy

## Files Included

- `app.py` - Main Gradio application
- `requirements.txt` - Python dependencies
- `image_database.xlsx` - Excel spreadsheet containing image URLs
- `README.md` - This file

## License

This application is provided as-is for demonstration purposes. The SigLIP2 model is provided by Google and subject to its own license terms.