a. Download yolo3.weight from this, and change the name to yolov3-608.weights.
b. python2 weight_to_onnx.py. To execute this script you must use python 2.7, and you will have a file named yolov3-608.onnx.
c1. python3 onnx_to_trt_1batch.py. If you only need to process one image each time, for example you only have one camera. Executing this script you need python 3.x, and you will have a file named yolov3-608.trt, which is the file we ultimately need.
c2. python3 onnx_to_trt_multibatch.py. If you need to process multiple images each time, for example you have multiple cameras. Executing this script you also need python 3.x, and you will have a file named yolov3-608.trt, which is the file we ultimately need. And the data accuracy is FP16, so the acceleration is more obvious.
d1.python3 trt_yolo3_module_1batch.py, if you choose c1
d2.python3 trt_yolo3_module_multibatch.py,if you choose c2. It detects 4 images at a time.
3. Import TensorRT_yolo3_module
This project has been packaged into class, so you can use it directly according import xx command.