Skip to content

Latest commit

 

History

History

csrnet

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
CMakeLists.txt
CMakeLists.txt
 
 
README.md
README.md
 
 
config.h
config.h
 
 
csrnet.cpp
csrnet.cpp
 
 
gen_wts.py
gen_wts.py
 
 
logging.h
logging.h
 
 
macros.h
macros.h
 
 

README.md

csrnet

The Pytorch implementation is leeyeehoo/CSRNet-pytorch.

This repo is a TensorRT implementation of CSRNet.

paper : CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes

Dev environment:

  • Ubuntu 22.04
  • TensorRT 8.6
  • OpenCV 4.5.4
  • CMake 3.24
  • GPU Driver 535.113.01
  • CUDA 12.2
  • RTX3080

how to run

1. generate csrnet engine
git clone https://github.com/leeyeehoo/CSRNet-pytorch.git
git clone https://github.com/wang-xinyu/tensorrtx.git
// copy gen_wts.py to CSRNet-pytorch
// generate wts file
python gen_wts.py
// csrnet wts will be generated in CSRNet-pytorch

2. build csrnet.engine
// mv CSRNet-pytorch/csrnet.engine to tensorrtx/csrnet
mv CSRNet-pytorch/csrnet.wts tensorrtx/csrnet
// build
mkdir build
cmake ..
make
sudo ./csrnet -s  ./csrnet.wts

Loading weights: ./csrnet.wts
build engine successfully : ./csrnet.engine

// download images https://github.com/wang-xinyu/tensorrtx/assets/46584679/46bc4def-e573-44ae-996d-5d68927c78ff and copy to images
sudo ./csrnet -d  ./images

// output e.g
// enqueueV2 time: 0.0323869s
// detect time:44ms
// people num :22.9101 write_path: ../images/data.jpg

result

inference people num: 22.9101