AI Powered Google-Lens

Image Similarity Engine (Google Lens Recreation with ResNet)

Overview

This project is an Image Similarity Search Engine that recreates Google Lens functionalities using a ResNet model. It allows users to find similar images based on a query image by performing feature extraction and similarity search.

Project Structure

.
├── Simmilarity_Search_Model.ipynb   # Jupyter Notebook for Image Similarity Search
├── Weights/                         # Pretrained ResNet Weights
├── Saved_Vector_database/           # Precomputed feature vectors for images
│   ├── filenames-caltech101.pickle  # Stores dataset image names
│   └── class_ids-caltech101.pickle  # Stores class names
├── requirements.txt                 # List of dependencies
└── Dataset/                         # Dataset containing image classes
    ├── Class1/
    ├── Class2/
    └── ...

Dataset: Caltech-101

The dataset consists of images of objects belonging to 101 categories, plus a background class. Each image is labeled with a single object. Details:

• 101 object categories + 1 background clutter class
• Images per category: 40 to 800
• Total images: ~9,000
• Image size: ~200-300 pixels per edge

Implementation Steps

1. Install Dependencies

Ensure you have Python 3.10 installed. Then, install required packages:

pip install -r requirements.txt

2. Setting Up the Dataset

• Organize the dataset as mentioned in the structure above.
• If the dataset is not present, it will be downloaded automatically.

3. Running the Similarity Search Model

Run the Jupyter Notebook to start the image search engine:

jupyter notebook Simmilarity_Search_Model.ipynb

4. How It Works

• Feature Extraction: The ResNet model extracts image features.
• Dimensionality Reduction: PCA reduces feature dimensions from 2048 to 150.
• Similarity Search: KNN finds the most similar images based on extracted features.
• Query Execution: A query image is compared against the dataset to retrieve similar images.

Google Lens Recreation: Key Features

• Accepts an image as input and finds similar images.
• Uses ResNet for feature extraction.
• Employs PCA to reduce dimensionality for better performance.
• Uses K-Nearest Neighbors (KNN) for fast retrieval of nearest neighbors.
• Efficient for large-scale image search applications.

Future Enhancements

• Optimize search time using faiss or Annoy.
• Improve accuracy by fine-tuning ResNet.
• Extend the system to work with real-time image uploads.

Contributing

Contributions are welcome! Feel free to raise issues or submit pull requests.

License

This project is open-source and available under the MIT License.