Upload 4 files

README.md

@@ -8,14 +8,39 @@ An image search application powered by Google's SigLIP2 model (`google/siglip2-s
 - 🖼️ 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. Encode all images in the database into embeddings
-2. Encode your text query into an embedding
-3. Find images with the highest similarity to your query
-4. Display the top matching results with similarity scores
+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
 
@@ -39,10 +64,6 @@ The app uses the SigLIP2 vision-language model to:
 
 This space uses the **google/siglip2-so400m-patch16-naflex** model, a state-of-the-art vision-language model from Google.
 
-## Dataset
-
-The app searches through a fixed collection of sample images from Unsplash covering various categories like nature, animals, cities, food, and more.
-
 ## Local Setup
 
 To run this locally:
@@ -52,13 +73,22 @@ 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` and `requirements.txt`
+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.

app.py

@@ -6,6 +6,8 @@ import numpy as np
 from typing import List, Tuple
 import requests
 from io import BytesIO
+import pandas as pd
+import os
 
 # Initialize model and processor
 MODEL_NAME = "google/siglip2-so400m-patch16-naflex"
@@ -16,24 +18,42 @@ processor = AutoProcessor.from_pretrained(MODEL_NAME)
 model = AutoModel.from_pretrained(MODEL_NAME).to(device)
 model.eval()
 
-# Fixed database of images (using sample images from various sources)
-IMAGE_DATABASE = [
-    "https://images.unsplash.com/photo-1506905925346-21bda4d32df4?w=400",  # Mountain landscape
-    "https://images.unsplash.com/photo-1518791841217-8f162f1e1131?w=400",  # Cat
-    "https://images.unsplash.com/photo-1552053831-71594a27632d?w=400",  # Dog
-    "https://images.unsplash.com/photo-1506748686214-e9df14d4d9d0?w=400",  # Beach sunset
-    "https://images.unsplash.com/photo-1469474968028-56623f02e42e?w=400",  # Nature/Forest
-    "https://images.unsplash.com/photo-1519681393784-d120267933ba?w=400",  # Mountains
-    "https://images.unsplash.com/photo-1504893524553-b855bce32c67?w=400",  # City skyline
-    "https://images.unsplash.com/photo-1541963463532-d68292c34b19?w=400",  # Flowers
-    "https://images.unsplash.com/photo-1488590528505-98d2b5aba04b?w=400",  # Technology/laptop
-    "https://images.unsplash.com/photo-1546069901-ba9599a7e63c?w=400",  # Food
-    "https://images.unsplash.com/photo-1511919884226-fd3cad34687c?w=400",  # Car
-    "https://images.unsplash.com/photo-1473186578172-c141e6798cf4?w=400",  # Person running
-    "https://images.unsplash.com/photo-1464822759023-fed622ff2c3b?w=400",  # Mountain peaks
-    "https://images.unsplash.com/photo-1470071459604-3b5ec3a7fe05?w=400",  # Nature scene
-    "https://images.unsplash.com/photo-1441974231531-c6227db76b6e?w=400",  # Forest path
-]
+# Load image URLs from Excel file
+def load_image_database(excel_file: str = "image_database.xlsx") -> List[str]:
+    """Load image URLs from Excel spreadsheet"""
+    if not os.path.exists(excel_file):
+        raise FileNotFoundError(
+            f"Image database file '{excel_file}' not found. "
+            f"Please create an Excel file with a column named 'url' containing image URLs."
+        )
+    
+    df = pd.read_excel(excel_file)
+    
+    # Look for a column named 'url', 'URL', 'image_url', or similar
+    url_column = None
+    for col in df.columns:
+        if col.lower() in ['url', 'image_url', 'image_urls', 'urls', 'link', 'image']:
+            url_column = col
+            break
+    
+    if url_column is None:
+        raise ValueError(
+            f"Could not find URL column in Excel file. "
+            f"Please use one of these column names: 'url', 'URL', 'image_url', 'urls', 'link', or 'image'. "
+            f"Found columns: {list(df.columns)}"
+        )
+    
+    # Extract URLs and remove any NaN values
+    urls = df[url_column].dropna().tolist()
+    
+    # Convert to strings and strip whitespace
+    urls = [str(url).strip() for url in urls]
+    
+    print(f"Loaded {len(urls)} image URLs from {excel_file}")
+    return urls
+
+# Load the image database from Excel
+IMAGE_DATABASE = load_image_database()
 
 # Cache for loaded images
 image_cache = {}
@@ -129,6 +149,8 @@ with gr.Blocks(title="Image Search with SigLIP2") as demo:
         Search through a collection of images using natural language queries!
         The model used is **google/siglip2-so400m-patch16-naflex**.
         
+        Image URLs are loaded from **image_database.xlsx**.
+        
         Try queries like:
         - "a cat"
         - "mountain landscape"
@@ -179,7 +201,7 @@ with gr.Blocks(title="Image Search with SigLIP2") as demo:
     gr.Markdown(
         """
         ---
-        **Note:** This demo uses a fixed set of sample images from Unsplash. 
+        **Note:** This demo uses images from the **image_database.xlsx** file. 
         The SigLIP2 model computes similarity between your text query and the images to find the best matches.
         """
     )

requirements.txt

@@ -4,3 +4,5 @@ transformers==4.46.0
 Pillow==10.1.0
 numpy==1.24.3
 requests==2.31.0
+pandas==2.1.1
+openpyxl==3.1.2