The visual stimuli will feed into the rotation cortical area via the built-in mapping rotator_z. This mapping encodes image orientation along the z-axis, so each z-layer represents a different rotation of the object. To isolate a particular orientation from the rotation stack, we use the rotation_select bar region. Both rotation and rotation_select project to rotation_out, which has z-depth 1, while rotation and rotation_select each have z-depth 10. Rotation_out activates only when it receives coincident input from both rotation and rotation_select. When the front voxel of rotation_select is activated, the activation propagates through its z-depth, effectively selecting one z-layer from rotation cortical area to be projected onto rotation_out at a time. This enables sequential, one-orientation-at-a-time processing of a multi-orientation z-stack, making it straightforward to train on multiple rotated versions of the target object.