https://github.com/denvercoder1/chess-piece-classifier

Demonstration of Tensorflow Lite for classifying images of chess pieces
https://github.com/denvercoder1/chess-piece-classifier

chess tensorflow tflite

Last synced: about 2 months ago
JSON representation

Demonstration of Tensorflow Lite for classifying images of chess pieces

• Host: GitHub
• URL: https://github.com/denvercoder1/chess-piece-classifier
• Owner: DenverCoder1
• License: mit
• Created: 2021-08-29T18:34:44.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2021-08-30T10:52:23.000Z (over 3 years ago)
• Last Synced: 2025-01-16T00:40:27.907Z (3 months ago)
• Topics: chess, tensorflow, tflite
• Language: JavaScript
• Homepage: https://denvercoder1.github.io/chess-piece-classifier/
• Size: 76.8 MB
• Stars: 1
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Chess Piece Classifier

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.

[![Peek 2021-08-29 21-18](https://user-images.githubusercontent.com/20955511/131261161-1fb7b30b-2250-4255-a6be-5444bf34925b.gif)](https://denvercoder1.github.io/chess-piece-classifier/)

## What is this?

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/

## Where is the data from?

The images found in [chess.tgz](data/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.

## How is the model made?

I used the [Tensorflow Image Classification Model Maker](https://www.tensorflow.org/lite/tutorials/model_maker_image_classification) to simplify the creation of the model.

![image](https://user-images.githubusercontent.com/20955511/131260911-44984a19-0177-447a-bb89-a115d151a6de.png)

The model was trained remotely using Google Colab and a copy of the Python notebook can be found [here](https://colab.research.google.com/drive/1AIsa43KHs2cywEfi3Xj5hJtrPI_Z7aiN?usp=sharing).

## What about the website?

The website uses methods from [Tensorflow.js](https://js.tensorflow.org/api/latest/) to load the model, pre-process the images, and generate predictions.

## Disclaimer

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.