Training data for the Imaginator of Beyond Thinking: Imagining in 360° for
Humanoid Visual Search.
An agent searching a 360° scene sees only a narrow field of view at a time.
Each row here teaches one step of that: given the views seen so far and an
instruction, write down the spatial layout of the scene — what is visible, and
what must lie in the directions nobody has looked at — and from that predict
where the target is.
