Sample 1: Running NN model (calculating right image) and computing the disparity map on OAK-D.
Outputs: [disparity map, right image, BBox].
Sample 2: Running NN model (calculating right image) on OAK-D.
Outputs: [left image, right image, BBox].
PS: During NN model compilation, VPU_NUMBER_OF_SHAVES and VPU_NUMBER_OF_CMX_SLICES use the same settings. (-VPU_NUMBER_OF_SHAVES 4 -VPU_NUMBER_OF_CMX_SLICES 4)
From Intel's official website, we know that one SHAVE is paired with one CMX.
From DepthAI's website, we learn that OAK-D contains 16 SHAVES and 20 CMX (each 128KB).
- Where might the resources of the additional 4 CMX be used?
- Samples 1 and 2 respectively consume 2.47 MiB and 2.5 MiB. Where might the additional 0.3 MiB be allocated?
- Does the OAK-D automatically adjust resource allocation? What are the aspects that undergo automatic adjustments?
