Code for training a small visual ANN to distinguish self rotation from object motion.
The natural scene images used in the ANN training were adapted from an online data base 'Panoramic high dynamic range images in diverse environments'. The code for preprocessing of the data can be found in the folder data_analysis.
The neural network models were standard CNNs constrained by spacial invariance, where filtered signals from different locations in space were summed uniformly. The code for model architectures and training can be found in the folder models.