/GuessMyDrawing-ios

Custom mlmodel trained on Google's quick draw data set.

Primary LanguageSwiftOtherNOASSERTION

Guess My Drawing

Guess My Drawing classifies drawn images, currently supporting 345 different categories. It displays the top-5 confidence categories. I trained the model on Google's Quick, Draw! dataset, and use iOS 11's CoreML framework to run the trained model.

Device Installation

  1. Download or clone this repository.
  2. Open the project file in Xcode.
  3. Build and install the application to your device.

Screenshots

Here is a GIF of it classifying a few drawings correctly: App

Training the Model

I am currently not releasing the details on how I trained the model. The architecture of the model is:

convolution (3x3 window, 64 filters) w/ relu
convolution (3x3 window, 32 filters) w/ relu
convolution (5x5 window, 32 filters) w/ relu
fully connected (1024 neurons) w/ relu
fully connected (512 neurons) w/ relu
fully connected (512 neurons) w/ relu
fully connected (345 neurons) w/ softmax