Skip to content

Yibin122/TensorRT-SMOKE

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

TensorRT-SMOKE

This repo deploys SMOKE using TensorRT for C++ inference.

Prerequisites

  • Install TensorRT 8.0
    # https://developer.nvidia.cn/nvidia-tensorrt-8x-download
    tar -xvzf TensorRT-8.0.3.4.Linux.x86_64-gnu.cuda-10.2.cudnn8.2.tar.gz
    # export TENSORRT_DIR=~/TensorRT-8.0.3.4
    git clone -b release/8.0 https://github.com/Yibin122/TensorRT.git  # MMCVModulatedDeformConv2d plugin
    cd TensorRT
    git submodule update --init --recursive
    mkdir -p build && cd build
    cmake .. -DTRT_LIB_DIR=~/TensorRT-8.0.3.4/lib -DTRT_OUT_DIR=`pwd`/out -DCUDA_VERSION=10.2  # e.g. CUDA 10.2
    make -j$(nproc)
    make install
  • MMDetection3D (v1.0.0)
    pip install torch==1.8.0 torchvision==0.9.0
    pip install mmcv-full==1.4.0
    pip install mmdet==2.19.0
    pip install mmsegmentation==0.20.0
    git clone -b v1.0.0rc0 https://gitee.com/open-mmlab/mmdetection3d.git
    cd mmdetection3d
    pip install -v -e .  # or "python setup.py develop"

Steps

  1. Export ONNX (first comment code in mmdet3d/models/dense_heads/smoke_mono3d_head.py)
    cd mmdetection3d
    python smoke_pth2onnx.py  # smoke_dla34.onnx
  2. Build engine
    cd TensorRT-8.0.3.4/bin
    # use your own model_path
    ./trtexec --workspace=4096 --onnx=${model_path}/smoke_dla34.onnx --saveEngine=${model_path}/smoke_dla34.trt8
  3. C++ inference
    cd TensorRT-SMOKE/
    mkdir build && cd build
    cmake ..
    make && ./smoke_trt
    kitti_000008_pred

TODO

  • MonoFlex

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published