Using the onboard VPU as a standlone accelerator

The VPU can be used as a standalone accelerator, without needing to launch and cameras or other sensors onboard the OAK-D.

An example is as follows:

import numpy as np
from oakutils.blobs.models.shave6 import GAUSSIAN_15X15
from oakutils import VPU


def main():
    vpu = VPU()
    vpu.reconfigure(GAUSSIAN_15X15)

    fps_buffer = deque(maxlen=30)
    while True:
        # generate some random data, then send to camera and wait for the result
        data = np.array(np.random.random((640, 480, 3)) * 255.0, dtype=np.uint8)
        _ = vpu.run(data)

if __name__ == "__main__":
    main()

This simple example will generate random data, send it to the VPU, and wait for the result. This simple abstraction can intake any blob file and run it on the VPU. The input to the run call MUST be a numpy array with the correct size buffer. If the buffer (overall size) of the data is not the expected size the VPU will crash.

The VPU can be reconfigured at any time, and the run call will wait for the result before returning. This allows for a simple abstraction to the VPU, and allows for the VPU to be used as a standalone accelerator.