SemantiSnap

A semantic image search tool powered by ResNet50. Upload a query image to find visually similar images from a folder, with similarity scores, summary visualization, and Excel/PPT export of results.

Semantic Image Search & Visual Summary Generator

A Python-based application that performs semantic image search using a pretrained ResNet50 model. Given a query image, it finds the most similar images from a folder using cosine similarity, and provides:

• Folder-level summary of results
• Visual comparison with the query
• Auto-generated Excel report with scores
• Optional PowerPoint presentation of the results

Features

Semantic search using deep features (ResNet50)
Cosine similarity-based retrieval
Query visualization with top-k similar results
Excel file with detailed similarity results
Folder-wise image match count summary
Easy folder scanning and feature database building

How It Works

1. Feature Extraction
   Uses a pretrained ResNet50 model. Features are extracted from the last pooling layer and normalized (L2 norm).

2. Database Building
   Scans the target folder recursively and extracts features for all valid images.

3. Semantic Search
   Given a query image, computes cosine similarity against all database features and returns the top-k matches.

4. Result Visualization

   • Displays the query and top similar images with similarity scores.
   • Creates an Excel file with two sheets:
   • Search Results (image-wise similarity)
   • Folder Summary (number of hits per folder)

Installation

git clone https://github.com/your-username/semantic-image-search.git
cd semantic-image-search
pip install -r requirements.txt