This is a simple image orientation detection model. It uses a CNN to identify upside down images.
It is trained on YouTube videos. This repository is intended to be educational as the model is not particularly complex or innovative, but instead minimally implemented.
Run ./download.sh
to download the dataset. yt-dlp is required to download the videos.
Run python run.py
to train the model. It will output the model to model.pt
in PyTorch format.