Hi guys,
I am working with NVIDIA Jetson TX2 board, which has an ARM processor and Pascal GPU architecture (the cpu is not Intel btw) running on Ubuntu 16.04 LTS. I need to run an object detection inference from OpenCVs DNN model zoo. As OpenCV (until 3.4.2) doesn't have CUDA implementation, I am restricted to the use of only CPU for DNN inference. Currently, on the cpu, the MobileNet-SSD Caffe implementation runs at around 200 ms. I was wondering if I compile OpenCV with Halide backend, would the inference be faster. Last I checked, the DNN Halide backend implementation was work in progress, and it performed clearly worse than the CPU implementation (although the CPU used was Intel). Any suggestions would be helpful.
Best Regards, s0r2637