This is a simple demonstration of the capabilities of a Tensorflow Lite image classification model.
It was created for a presentation in an Introduction to AI course in August 2021.
This is a basic web interface that uses a Tensorflow Lite image classifier to classify chess pieces.
View the demo here: https://denvercoder1.github.io/chess-piece-classifier/
The images found in chess.tgz were created by taking 1-minute videos of each of the six types of pieces and splitting them into frames using VLC. The videos were filmed at 30 frames per second and 1 frame for every 50 frames was extracted from each video.
I used the Tensorflow Image Classification Model Maker to simplify the creation of the model.
The model was trained remotely using Google Colab and a copy of the Python notebook can be found here.
The website uses methods from Tensorflow.js to load the model, pre-process the images, and generate predictions.
This is just a simple demonstration and it was trained only on a small dataset consiting of only photos of one chess set.
To avoid overfitting, it is recommended to try with a larger dataset and perform augmentation to generate more varied data.