An implementation of paper Detecting anomalous events in videos by learning deep representations of appearance and motion on python, opencv and tensorflow. This paper uses the stacked denoising autoencoder for the the feature training on the appearance and motion flow features as input for different window sizes and using multiple SVM as a weighted single classifier this is work under progress if anyone can contribute I would be glad to work.
Most of the files contains the script and details in the files. Once scpit splices the imges of different size for apperance model: windows size - 15x15, 18x18, 20x20 Denoising auto encoder file to train the model from the pickle file where you have created the dataset from the images.
