Edge end development: At present, we have adopted object detection algorithm in Jetson Xavier NX, accelerated by TensorRT, converted to WTS through PT, and then converted to TensorRT engine file (.engine) on the edge side; Can this scheme achieve DLA acceleration?
Reference: tensorrtx/yolov5 at master · wang-xinyu/tensorrtx · GitHub
config.default_device_type = trt.DeviceType.DLA
Could you point in repo, where you are setting the device to be used as DLA? Note that, DLA may not support all the layers in the network. Please set GPU fallback flag as well to make sure non DLA supported layers fall back to GPU.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.