ChessApp

Deep Learning applicated to a computer vision project.
ChessApp is an Android application for chess board detection and pieces recognition, using a neural network for classification trained on home-made dataset and tested with different techniques and approaches.
The application also creates a virtual representation of the board allowing the user to visualize the classification results.
Created in collaboration with Simone Mattioli

Datasets

We created our own dataset by collenting chess board images, crop each box and label them.
We have a small dataset made of the 12.500 original images and a larger one made by 110.600 images.
All the datasets can be found here.

Models

We generated up to 13 models using different approaches and training techniques.

13 classes approach (one model)

### 2 + 7 classes approach (two models) 2 classes:

7 classes:

All the checkpoints can be found here.