Vehicle Image Similarity Search using Deep Learning and PySpark

Overview

This project focuses on finding similar vehicle images using deep learning and PySpark. It involves extracting image embeddings using a pre-trained ResNet-18 model, storing these embeddings, and performing a similarity search using Locality-Sensitive Hashing (LSH) in PySpark. The goal is to identify the most similar vehicles based on their image features.

Data Source: The project uses the Stanford Cars Dataset, which contains images of cars from different classes..

Table of Contents

• Project Features
• Project Architecture
• Technologies Used
• Running the Application
• Project Structure
• Contact

Project Features

• Image Embedding Extraction: Uses a pre-trained ResNet-18 model to extract feature vectors (embeddings) from vehicle images.
• Locality-Sensitive Hashing (LSH): Implements LSH in PySpark to efficiently find similar images based on their embeddings.
• Scalable Similarity Searchr: Handles large datasets by leveraging PySpark's distributed computing capabilities.
• Show Results: Displays the most similar images for a given query image.

Project Architecture

Below is the high-level architecture of the project, including image acquisition, preprocessing, digit recognition, solution, and visualization stages.

1. Image Preprocessing: Resizes and normalizes images to prepare them for feature extraction.
2. Feature Extraction:: Uses a pre-trained ResNet-18 model to extract 512-dimensional feature vectors from the images.
3. LSH Indexing: Applies Locality-Sensitive Hashing (LSH) to index the feature vectors for efficient similarity search.
4. Similarity Search: Queries the LSH index to find the most similar images for a given input image.
5. Results Visualization: Displays the query image and its nearest neighbors.

Technologies Used

• Python: Core programming language.
• PyTorch: For loading and using the pre-trained ResNet-18 model.
• PySpark: For distributed computing and implementing LSH.
• Pillow/Pandas: For image handling and manipulation and storage.

Running the Application

1. Clone the Repository:

   • First, clone the repository to your local machine:

     git clone https://github.com/Carlos93U/Vec2Car.git
     cd Vec2Car

2. Create and Activate Virtual Environment:

   • It is recommended to create a virtual environment to manage the dependencies:

     python -m venv env
     source env/bin/activate  # Linux/Mac
     .\env\Scripts\activate  # Windows

3. Run first cell on notebook to install dependencies:

   • Install the required dependencies running first cell on notebook:

     !pip install pandas
     !pip install pyspark
     !pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cpu
     !pip3 install tqdm

4. Run all cells on notebook:

   • Run all cells on notebook to extract embeddings and perform similarity search:

Project Structure

The following is an overview of the main folders and files within the project.

Vec2Car/
├── data/                      # Data files and images
│   ├── car_dataset_test.csv
│   ├── car_dataset_train.csv
│   ├── input_data_vectors.csv
│   ├── cars_test/
│   ├── cars_train/
│   └── input_images_cnn/
│
├── notebooks/                  # Notebook for data preprocessing and model training
├── docs/                       # Project documentation
└── README.md                   # Project documentation

Contact

For any questions or feedback, please contact:

Juan C. Huillcas
Email: huillcas.juan3@gmail.com
LinkedIn: Juan C. Huillcas

---