QuickDraw

This project is to train a model to recognize drawing.

Prerequisites

This page contains the list of prerequisites for canvas

• Ubuntu

    sudo apt-get install build-essential libcairo2-dev libpango1.0-dev libjpeg-dev libgif-dev librsvg2-dev
  

Installation

npm install

Benchmark

The benchmark contains python and js code to see how faster it is to convert strokes to images. Resizing an image in js is faster when using ctx.translate instead of using a second canvas for cropping.

script

The script files allowd to download the quickdraw ndjson files, to generate the labels file and to train the model.

Download the ndjson files

• Download the quickdraw dataset files

  ./script download

• Generate label types

  ./script create <training | validation | testing >

• Train the model

  ./script train

See in action

In this repository the model is deployed server side. But in can be possible to deploy it directly in the browser.

launch the server

ts-node server/server.ts

launch the client

npx parcel client/index.html