Crate plato_vision_jepa 

plato-vision-jepa

Vision JEPA for the PLATO nervous system. Processes camera frames into structured room state vectors suitable for downstream nervous-system tiles.

Structs

RoomVisionState

16-dimensional vision state vector for a room.

VisionDeadband

Deadband filter — only pass frames whose histogram diff exceeds a threshold.

VisionTile

Structured room state produced by the vision JEPA from a single frame.

Functions

compute_frame_diff

Compute histogram intersection distance between two frames. Returns a value in [0, 1] where 0 = identical, 1 = maximally different.

compute_motion_vector

Compute average motion vector from tracked point positions.

extract_region_states

Divide an image grid into 4 quadrants and return average intensity of each normalized to [0, 1].

is_significant_change

Determine whether a frame diff exceeds the significance threshold.

vision_state_to_tile

Convert a RoomVisionState into a VisionTile.