This repository maintains the code for 828j project. It is divided into two parts: data sampling, and training/nets
./data_sampling directory contains code for data sampling where high motion aware tuples are sampled. More details here
./net directory contains a Two-Stream network built upon AlexNet. More details here
unsupervised_train.py and supervised_train.py use the two-stream network and ./data_sampling/tuple_loader to train the network during unsupervised (aka self-supervised) and supervised mode respectively.