Implementation of Quickdraw - an online game developed by Google
Branch: master
Clone or download
Latest commit cbba7f4 Jan 31, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
all_images upload Jan 31, 2019
data upload Jan 31, 2019
demo upload Jan 31, 2019
images upload Jan 31, 2019
src upload Jan 31, 2019
tensorboard upload Jan 31, 2019
trained_models upload Jan 31, 2019
LICENSE upload Jan 31, 2019
README.md upload Jan 31, 2019
camera_app.py upload Jan 31, 2019
draw.py upload Jan 31, 2019
painting_app.py upload Jan 31, 2019
train.py upload Jan 31, 2019

README.md

[PYTHON] QuickDraw

Introduction

Here is my python source code for QuickDraw - an online game developed by google. with my code, you could:

  • Run an app which you could draw in front of a camera (If you use laptop, your webcam will be used by default)
  • Run an app which you could draw on a canvas

Camera app

In order to use this app, you need a pen (or any object) with blue, red or green color. When the pen (object) appears in front of camera, it will be catched and highlighted by an yellow circle. When you are ready for drawing, you need to press space button. When you want to stop drawing, press space again Below is the demo by running the sript camera_app.py:


Camera app demo

Drawing app

The script and demo will be released soon

Dataset

The dataset used for training my model could be found at [Quick Draw dataset] https://console.cloud.google.com/storage/browser/quickdraw_dataset/sketchrnn. Here I only picked up 20 files for 20 categories

Categories:

The table below shows 20 categories my model used:

apple book bowtie candle
cloud cup door envelope
eyeglasses guitar hammer hat
ice cream leaf scissors star
t-shirt pants lightning tree

Trained models

You could find my trained model at trained_models/whole_model_quickdraw

Training

You need to download npz files corresponding to 20 classes my model used and store them in folder data. If you want to train your model with different list of categories, you only need to change the constant CLASSES at src/config.py and download necessary npz files. Then you could simply run python3 train.py

Experiments:

For each class, I take the first 10000 images, and then split them to training and test sets with ratio 8:2. The training/test loss/accuracy curves for the experiment are shown below:

Requirements

  • python 3.6
  • cv2
  • pytorch
  • numpy