Hi! I’m using a retrained ssd_mobilenet_v1 in jetson-inference. My network has 2 classes (+1 unmarked). In the frame from the camera, I define a class 1 object. By its bbox, I should allocate the ROI in which I should find a class 2 object. But since the image from the camera was received by the command " img, width, height = camera.CaptureRGBA(zeroCopy = True)", I can’t do anything with it. The conversion by the cudaToNumpy() function greatly increases the running time of the algorithm from ~40 ms to ~ 120 ms. Is there a way to directly get the swarm area from img?
(jetson AGX Xavier, linux ubuntu 18.04 LTS, python 3.6.9, tensorflow 1.14.0, tensorrt 6.0.1…)