Now, just support Ubuntu, windows will be soon.
Support yolov3 yolov3-tiny yolov4 yolov4-tiny,even your custom cfg networ
if you train from darknet(AB), it usually can support.
Don't need onnx, directly transport .cfg and .weights to Tensorrt engine
this project borrow from Deepstream and wang-xinyu/tensorrtx
1. clone.
2.set CMakeList.txt tensorrt path, opencv path.
3.main.cpp, set diffenrt cfg and weights
4.set .cfg input_w and input_h,due to tensorrt upsample , input_w shuld equal input_h
5.copy .cfg and .weights file to folder
6.mkdir build.
7.cd build && cmake .. && make
7.run ./yolo -s to build yolo engine
7.run ./yolo -d to start detect
- FP16/FP32 can be selected by the macro
USE_FP16