https://github.com/hasnainroopawalla/scribblenet

A Python library that classifies hand-drawn doodles into distinct classes (trained on the QuickDraw dataset by Google).
https://github.com/hasnainroopawalla/scribblenet

android cnn doodle-classifier flask javascript keras python restful-api tensorflow

Last synced: 5 days ago
JSON representation

A Python library that classifies hand-drawn doodles into distinct classes (trained on the QuickDraw dataset by Google).

• Host: GitHub
• URL: https://github.com/hasnainroopawalla/scribblenet
• Owner: hasnainroopawalla
• License: mit
• Created: 2020-01-13T15:35:58.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2022-01-04T20:51:06.000Z (about 3 years ago)
• Last Synced: 2023-03-11T05:28:49.339Z (almost 2 years ago)
• Topics: android, cnn, doodle-classifier, flask, javascript, keras, python, restful-api, tensorflow
• Language: Jupyter Notebook
• Homepage: https://www.hasnainr.com/projects/scribblenet.html#demonstration
• Size: 52.3 MB
• Stars: 3
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
ScribbleNet


[![Linting](https://github.com/hasnainroopawalla/ScribbleNet/actions/workflows/linting.yml/badge.svg)](https://github.com/hasnainroopawalla/ScribbleNet/actions/workflows/linting.yml)

[![Python 3.9](https://img.shields.io/badge/python-3.9-blue.svg)](https://www.python.org/downloads/release/python-390/)

![Status](https://img.shields.io/badge/status-stable-green.svg)

Check out [the demo](https://www.hasnainr.com/projects/scribblenet.html#demonstration)!

ScribbleNet is Python library that classifies hand-drawn doodles into distinct classes (trained on the [QuickDraw dataset by Google](https://github.com/googlecreativelab/quickdraw-dataset)).

This Python package is bundled as a REST API using [Flask](https://flask.palletsprojects.com/en/2.0.x/) and deployed on [Heroku](https://www.heroku.com/).

The front-end is built using vanilla JavaScript, HTML and CSS.

Download the Android App here ([Google Play](https://play.google.com/store/apps/details?id=doodle.classifier)).

>  The repository has recently been refactored and thus, the Android application currently doesn't work.

## 📝 Table of Contents

- [Getting Started](#gettingstarted)

- [Machine Learning Stuff](#ml)

## 🏁 Getting Started 

[Working demo](https://www.hasnainr.com/projects/scribblenet.html#demonstration)

To run locally:

1.  ```

    $ git clone https://github.com/hasnainroopawalla/ScribbleNet.git

    $ pip install -r requirements.txt

    $ python3 scribblenet/server.py

    ```

2. Navigate to `127.0.0.1:5000` in your browser.

## 🧮 Machine Learning Stuff 

### Pre-processing

- Two pre-processing pipelines (for training, predicting on external data) have been created. Refer to `scribblenet.preprocessor.PreProcessor` ([source](https://github.com/hasnainroopawalla/ScribbleNet/blob/b645e1c1299784faebbce4f7efbbdd67758bae0b/scribblenet/preprocessing/preprocessor.py#L19)).

### Training

- A `Jupyter` notebook for training the model on the Quickdraw Dataset can be found [here](https://github.com/hasnainroopawalla/ScribbleNet/blob/master/scribblenet/ml/training.ipynb).

- The notebook uses utility methods from the `scribblenet` package for loading and preprocessing the data.

- The `load_classes()` method accepts `100` or `all` as an argument to indicate if 100 classes or all 345 classes should be loaded.

- The `load_dataset()` method accepts `num_samples_per_class` as an argument to indicate how many samples of each class should be loaded.