trellixvulnteam / tensorrt-centernet-3d_idzi Goto Github PK

View Code? Open in Web Editor NEW

tensorrt5, centernet, 3d detection, kitti, centerface, deform conv, int8,

License: MIT License

C++ 70.20% Python 10.82% C 0.84% Cuda 14.75% CMake 2.26% Dockerfile 0.41% SWIG 0.73%

tensorrt-centernet-3d_idzi's Introduction

TensorRT-CenterNet-3D

Most parts of this Repo are based on CaoWGG, Thanks for that great work, and I didn't find an open-sourced 3d-detection based on CenterNet (2020.06). So I followed the work of Cao and test the 3D detection based on CenterNet with the acceleration of TensorRT.

1. Enviroments

NVIDIA Quadro M4000

pytorch 1.1
ubuntu 16.04
TensorRT 5.0
onnx-tensorrt v5.0
cuda 10.1

2. Performance

3. How to use

download the ddd_3dop.pth from here
Convert CenterNet-3D model to onnx. See here for details, Use netron to observe whether the output of the converted onnx model is (('hm', 3), ('dep', 1), ('rot', 8), ('dim', 3), ('wh', 2), ('reg', 2))
set the OPENCV path in ./src/CMakeLists.txt ,set the TENSORRT path in ./CMakeLists.txt
uncomment the configs of 3-D model in ./include/ctdetConfig.h and comment others.
build

git clone https://github.com/Qjizhi/TensorRT-CenterNet-3D.git
cd TensorRT-CenterNet-3D
mkdir build
cd build && cmake .. && make
cd ..

build the engie and do the inference

./buildEngine -i model/ddd_3dop.onnx -o model/ddd_3dop.engine
./runDet -e model/ddd_3dop.engine -i 000292.png -c test.h264

4. Good to know

This codes are just for testing and will not be well maintained, for any suggestions and improvements, please propose an issue.
With the help of TensorRT, the inference time has been shortened from 0.227s to 98.89ms.
The functions of previous Repo are still remained, for exmaple, centerface...