See https://github.com/SensorsINI/DHP19 for code to generate the constant count accumulated DVS frames that we used for training a Convolutional Neural Network (CNN), as well as the keras pretrained CNN model that we trained on 2 camera views (camera 2 and 3).