darknet.weights --> onnx.onnx --> tensorrt.trt
sample from TensorRT-5.x.x.x/samples/python/yolov3_onnx/
device:
nvidia tx2
environment:
ubuntu18.04
tensorrt5.0.6.3
cuda10.0
cudnn7.3.1
set labels: line14,line19|data_processing.py
set output_layers: line723|yolov3_to_onnx.py
set batch_size: batch|xxx.cfg, line62|yolov3_to_onnx.py, line140|onnx_to_tensorrt.py
set input_size: input|xxx.cfg, line63|yolov3_to_onnx.py, line139|onnx_to_tensorrt.py
set mode: line141|onnx_to_tensorrt.py
input batch size inference time/ms
416x416 1 9.5
16 8.75
32 8.5
608x608 1 20
16 18.1
32 18