Skribix

Skribix focuses on hand-drawn sketch recognition. It leverages machine learning and deep learning techniques to classify sketches into various object categories. The project is designed to provide an easy-to-use interface for training and testing models on sketch datasets.

Make your sketches here - Skribix - Handdrawn sketch recognition

Versioning

• skribix_v1: Utilizes 250 classes of images for image recognition.
• skribix_v2: A scaled-down version with 15 handpicked classes of images for improved accuracy and performance.

Features

• Classification of hand-drawn sketches into multiple object categories.
• Implementation of various machine learning algorithms:
  • KNN
  • SVM
  • Decision Trees
  • Naive Bayes
  • ANN
  • CNN
• Robust feature extraction techniques for enhanced accuracy.
• Easy-to-use interface for training and testing models.
• Modular structure for extensibility.

Installation

1. Clone the repository:

   git clone https://github.com/Sahil-1918912/Skribix.git

2. Navigate to the project directory:

   cd Skribix-2

3. Install the required dependencies:

   pip install -r requirements.txt

Project Structure

    skribix/
    ├── backend_api   
    │   ├── api.ipynb
    │   ├── api.py
    │   └── test.ipynb         
    ├── frontend   
    │   ├── public
    │   └── src
    │        ├── assets
    │        │      └── react.svg
    │        ├── App.css
    │        ├── App.jsx
    │        ├── DrawingTool.jsx
    │        ├── index.css
    │        └── main.jsx
    ├── skribix_v1 
    │   ├── feature extraction   
    │   └── models  
    ├── skribix_v2   
    │   ├── ann model  
    │   ├── bayesian_model
    │   ├── best model
    │   ├── clustering
    │   ├── cnn model 
    │   ├── edge_detection
    │   ├── feature extraction
    │   ├── feature_extraction_smooth
    │   ├── ImageDataGenerator
    │   ├── knn
    │   ├── knn model
    │   ├── sketches
    │   └── svm model
    ├── web page 
    │   └── index.html               
    ├── .gitignore         
    ├── LICENSE                  
    ├── README.md
    └── requirements.txt

Usage

Backend

1. Start the backend API:

   python backend_api/api.py

2. Test the backend using the provided Jupyter Notebook:

   jupyter notebook backend_api/test.ipynb

Frontend

1. Navigate to the frontend directory:

   cd frontend

2. Install frontend dependencies:

   npm install

3. Start the development server:

   npm run dev

Training Models

1. Train a model using one of the supported algorithms:

   python skribix_v2/<algorithm_directory>/train.py --data <path_to_dataset>

   Replace <algorithm_directory> with the desired algorithm directory.

Testing Models

1. Test a trained model:

   python skribix_v2/<algorithm_directory>/test.py --model <path_to_model> --data <path_to_test_data>

   Replace <algorithm_directory> with the desired algorithm directory.

Models Used

• PCA + Naive Bayes
• PCA + KNN
• ANN (Artificial Neural Network)
• CNN (Convolutional Neural Network)
• K-Means + KNN
• Mean-Shift Clustering + KNN
• GMM + KNN
• SVM
• K-Means + SVM
• PCA + SVM

Dataset

You can use publicly available sketch datasets such as:

1. How Do Humans Sketch Objects?

2. Hand-picked Dataset: skribix_v2/sketches
   Custom Feature Extraction: The project includes cutomized features extracted manuall and stored in .npy file format. These files contain preprocessed data specifically tailored for training and testing particular models.
   .npy file directories -

• skribix_v2/feature extraction
• skribix_v2/feature_extraction_smooth
• skribix_v2/edge_detection
• skribix_v2/IMageDataGenerator

References

• How Do Humans Sketch Objects?
• 2012_siggraph_classifysketch_paper
• Scikit-learn documentation for machine learning algorithms.
• TensorFlow for deep learning models.

License

This project is licensed under the MIT License. See the LICENSE file for details.