An introduction to the step by step video tutorial series on making an image classifier for doodle recognition.
Using TensorFlow, we will learn how to do image classification to make a game like “Quick, Draw!“.
It is an online game that challenges players to draw a doodle and then artificial intelligence guesses what the drawings represent.
This tutorial series shows the whole process of developing this game.
1. Video
To get a quick insight into the game we are going to make, watch this video:
2. Tutorial Info
Using Phaser 2 framework for making HTML5 games and Tensorflow.js library for implementing machine learning in Javascript, we will create a complete HTML5 Doodle Predictor game that runs directly in the browser and recognizes doodles.
To perform image recognition, the game uses Artificial Intelligence (AI) powered by Machine Learning (ML) and Convolutional Neural Network (CNN).
To avoid loading big data files, the model will be trained on a small subset of the Quick Draw Dataset using only 500 images per each category.
This means that the output predictions will not always be correct because the model is not trained for each possible case.
In order to get better and more accurate predictions from an image classifier, its convolution neural network should be trained using much more images. But for learning purposes this is quite enough.
3. Tutorial Topics
This tutorial series is divided into the following 8 parts:
- Part 1: project setup
- Part 2: getting data
- Part 3: building a convolutional neural network model
- Part 4: training the model
- Part 5: predicting a batch of samples
- Part 6: drawing doodles
- Part 7: recognizing doodles
- Part 8: adding more doodle categories
Each part contains a video and a fully commented source code so you can easily follow the game development step by step.
4. Playing Doodle Predictor
To get even more familiar with the game we’re going to make, here you can play it!
So this is the fully completed Doodle Predictor game that can recognize 3 different doodles: a car, a snowman, and a fish.
The games starts with loading the datasets and then you need to train and test the model by pressing the control buttons.
Once you are satisfied with the trained model, draw a doodle and check if the given predictions from the AI agent are correct. Enjoy!
5. Source Code
The complete source code for Doodle Predictor is available on Github:
https://github.com/ssusnic/Machine-Learning-Doodle-Recognition
Next time, we will start programming the game. So stay tuned!