Darknet for Jetson Nano

Hi All,

I would like to run tiny-yolo using darknet in half precision CUDNN. How can I achieve this without using tensor rt, because I am getting almost same latency for both darknet and tensor rt in fp32 mode.

Hi ahmad.nn102, I am not sure that the DarkNet framework natively supports FP16 mode. For FP16 you would probably want to use TensorRT, which could be significantly faster.

You could try the TensorRT-enabled YOLOv3 benchmark from here, and also a YOLO sample now included with TensorRT. You can find it on your Nano at /usr/src/tensorrt/samples/python/yolov3_onnx

Thanks,