Code for my Insight project, AutoDetect. This project is about finding anomalies in videos.
This repo contains two approaches: supervised and unsupervised/self-supervised. The code for Supervised is pure PyTorch, whereas the Unsupervised codebase is a mix of numpy/pytorch for the baseline (simple Frame2Frame similarity) and piggybacks on Pix2Pix for the GAN.
Please check my slide deck to see the results of each model.