Quickdraw 10 CNN Classifier

Demo

👉 https://quickdraw-10-classification.herokuapp.com/

Introduction

This project is demo project that getting hand-drawing image and try to classification which kind of image of input.

It is machine learning project to classify Google Quickdraw image just only 10 classes with convolutional neural networks using Keras and Tensorflow.
And also serve hand-writting website with Flask

Website Usage

1. Compose Docker

docker-compose up

2. Start http://localhost:8000

Basicly, Flask website view download model from model folder which I've provided pre-trained model model-v1.h5 and serve to website.

You can train new model with docker-compose up -f docker-compose.train.yml to start Jupyter notebook and customize code and run by yourself then save to model directory and replace the old model if you want to or change the code in web folder for better choice.

Model Usage

Training Data Preparation

Enter these command to automatic download training dataset

chmod +x ./data-download.sh
./data-download.sh

# Or
docker-compose exec quickdraw ./data-download.sh

Jupyter Usage

docker-compose up -f docker-compose.train.yml

Heroku Deployment

heroku container:login
heroku create
heroku container:push web
heroku open

License

MIT © Kosate Limpongsa