river20140407/deeplearning

Deep Learning sample projects with PyTorch

Deep Learning example projects

This repository contains multiple projects I implemented as a preparation for my bachelors thesis on the Bern University of Applied Sciences.

All projects use PyTorch and are located in the notebooks/ subdirectory. Python dependencies are saved in requirements.txt. The docker/ directory contains a Dockerfile to build a docker container containing PyTorch, CUDA and jupyter notebook.

MNIST

An implementation of the MNIST handwritten digit classification dataset.

Music

Detect the genre of a song. Use music_prepare.ipybn to generate a dataset from a local music library, then music.ipynb to train the classifier.

Icons

Detects 2D shapes (tetris stones) in real time from a webcam. This is a multi-label classifier because it can detect multiple shapes under the webcam.

Shapes3D

This is a unfinished project. It currently downloads and extracts 3D model files (*.stl) from thingiverse. The next step would be to generate renderings from different angles of the 3D models and then use these as input for the network.

One idea for a classifier would be to give the network three images of the same model from different angles and as a second input another image from the same or from another model. Then the network has to decide if the second input shows the same image as the three sample images.

Tetris

A very early attempt to create a Tetris bot. Based on code by Kevin Chabowski https://gist.github.com/silvasur/565419/d9de6a84e7da000797ac681976442073045c74a4, adjusted to work inside a jupyter notebook. Not bot code yet.