This repository has two folders: one for the Spatio Temporal Autoencoder for abonormal event classification and another for time series analysis on pose estimation. Both tasks revove around the Avenue dataset.
This model is based on the paper at the link https://arxiv.org/pdf/1701.01546.pdf However, some changes were made to the model and loss fucntions. Specifically, I changed the loss function from mean squared error into categorial cross-entropy as I thought it was appropriate since we are prediciting categories of abnormal events.
First the model is described as the one in the paper:
After preprocessing the data running python preprocessing.py /source/to/dataset/ fps, this model was then trained on the Avenue dataset for 32 epochs by running python train_model.py
