Models - CNN, MLP, GAN, RF
This project aims at predicting scribbles from canvas and also using GAN generated images.
Idea was inspired from Kaggle's competition - https://www.kaggle.com/c/quickdraw-doodle-recognition/overview
-
Trained a GAN model on few categories of objects from Quick Draw dataset. Save generated images in directory.(scribble_generation_gan.py)
-
Created script to download the data directly from the quick draw dataset website using download_data.py.
-
Built a training model using a ConvNet and MLP Model train.py.
-
Developed prediction model server.py which takes input from either canvas (from webapp) or GAN category dropdown (use saved images) and classify the scribbles among 7 categories.
-
Presented the results on webapplication.
- ML Libraries - Tensorflow, Keras, Scipy, Python
- Webapp - Electron JS, HTML, JQuery and Flask App.
NOTE: This project works in Tensorflow 2.x with v1 compatibility.
Web Application
British Columbia Artificial Intelligence Showcase Poster
GAN Generated sample
GAN Loss
