Table 2.

Specification of datasets with dot-defined environments. T indicates optic flow that contains only translation, and $T+R$ indicates optic flow with both translational and rotational components. Training set is used to learn optic flow templates, while test set is used to evaluate performance on novel patterns. Uniform indicates sampling from the uniform random distribution with the specified endpoints. Azimuth and elevation angles of 0 correspond to straight-ahead heading.

Dataset	Description	Size (Num Videos)	Independent Variables
3D dot cloud (T)	Translational self-motion through 3D dot cloud along random heading directions	Train: 500 Test: 250	Heading azimuth and elevation: $\mathrm{Uniform}(-45°,45°)$
Ground plane (T)	Translational self-motion over a ground plane with random heading directions	Train: 500 Test: 250	Heading azimuth and elevation: $\mathrm{Uniform}(-45°,45°)$
3D dot cloud ($T+R$)	Self-motion through 3D dot cloud along random 2D heading directions with varying amounts of pitch, yaw, roll rotation	Train: 1000 Test: 500	Heading azimuth and elevation: $\mathrm{Uniform}(-45°,45°)$ Rotation speed: $\mathrm{Uniform}(1,10)°/$s