Figure 3.

Schematic representation of the motion energy model (based on Nishimoto et al. 2011), described under Materials and Methods, Data Analysis, Spatiotemporal Motion Energy Model. In short, input stimuli (movies) are run through a bank of quadrature pairs of Gabor filters, each with a certain spatiotemporal frequency and orientation and located on a grid covering the stimulus. The output of each pair is then squared and summed to give a physiologically plausible measure of motion energy. The end result is finally obtained by taking the square root to model a compressive nonlinearity. This final output is calculated for each of the spatiotemporal frequencies and locations covered by the bank of Gabor filters and standardized per filter across frames. In a next step, the neural response is predicted as a linear combination of those standardized outputs.