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
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
The script and demo will be released soon
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
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 |
You could find my trained model at trained_models/whole_model_quickdraw
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
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:
- python 3.6
- cv2
- pytorch
- numpy