Object Detection and Retrieval System

This repository contains a Flask-based web application designed for detecting objects in images and retrieving visually similar images from a dataset using advanced deep learning and computer vision techniques. This project was implemented in 2020-2021.

Features

• Object Detection: Utilizes YOLOv3 and Detectron2 for detecting objects in uploaded or URL-based images.
• Image Embedding: Extracts embeddings using NASNetLarge and other deep learning models.
• Content-Based Image Retrieval (CBIR): Retrieves visually similar images from the dataset using embeddings and re-ranking methods.
• Re-Ranking: Refines retrieval results based on dominant color and texture similarity.
• Elasticsearch Integration: Manages image indexing and metadata storage.

Technologies Used

• Framework: Flask
• Deep Learning: Keras (NASNetLarge, InceptionResNetV2,...)
• Search Engine: Elasticsearch
• Object Detection: YOLOv3, Detectron2
• Reranking: Color and texture-based methods
• Additional Libraries: NumPy, OpenCV, skimage

Prerequisites

• Python 3.8 or higher
• Elasticsearch installed and running
• Required Python packages (install via requirements.txt)

Installation

1. Clone the repository:

   git clone https://github.com/nematollahsaeidi/Object-Detection-and-Retrieval-System.git
   cd Object-Detection-and-Retrieval-System

2. Install dependencies:

   pip install -r requirements.txt

3. Configure the application:

   • Update the cfg/config.cfg file with your settings (proxy, directories, etc.).
   • Ensure Elasticsearch is configured and accessible.

4. Start the Flask application:

   python app.py

Usage

1. Uploading Images

• Access the upload page via http://localhost:5002/.
• Upload an image or provide a URL for analysis.

2. Object Detection

• Detected objects and bounding boxes are displayed.

3. Image Retrieval

• Retrieves visually similar images based on embeddings and reranking methods (color and texture).

4. Displaying Results

• View retrieved images and their details on the web interface.

API Endpoints

• /receive_image: Accepts a URL-based image for detection and retrieval.
• /visual_search: Retrieves similar images based on uploaded or URL images.
• /display/<filename>: Displays a specific image.
• /displays/<filename>: Displays images from a category.

Configuration

Key settings in cfg/config.cfg:

• proxy: Proxy settings for external requests.
• request_dir: Directory for temporary request storage.
• dataset_folder: Path to the dataset directory.
• upload_folder: Directory for uploaded files.