Feature: add modified yolop tensorrt implementation (wang-xinyu#1129)

* add simplified Logger class

* add modified yolop tensorrtx implementation

* add gen_wts.py

* add CMakeLists.txt

* add yolop tensorrtx README.md

* fix yolop spelling

* simplify yolop code

* fix yolop README.md

Co-authored-by: aliceint(zhanglj) <[email protected]>

Author: ausk

yolop/CMakeLists.txt

@@ -0,0 +1,35 @@
+cmake_minimum_required(VERSION 2.6)
+
+project(yolop)
+
+add_definitions(-std=c++11)
+
+option(CUDA_USE_STATIC_CUDA_RUNTIME OFF)
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_BUILD_TYPE Release)
+
+find_package(CUDA  REQUIRED)
+
+include_directories(${PROJECT_SOURCE_DIR}/include)
+
+find_package(OpenCV REQUIRED)
+include_directories(${OpenCV_INCLUDE_DIRS})
+
+# cuda
+include_directories(/usr/local/cuda-10.2/include)
+link_directories(/usr/local/cuda-10.2/lib64)
+# tensorrt
+include_directories(/usr/include/aarch64-linux-gnu/)
+link_directories(/usr/lib/aarch64-linux-gnu/)
+
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -Ofast -Wfatal-errors -D_MWAITXINTRIN_H_INCLUDED")
+
+# to generate plugins
+cuda_add_library(myplugins SHARED ${PROJECT_SOURCE_DIR}/yololayer.cu)
+target_link_libraries(myplugins nvinfer cudart)
+
+# to generate trt and test image dir
+add_executable(yolop ${PROJECT_SOURCE_DIR}/yolop.cpp)
+target_link_libraries(yolop nvinfer cudart myplugins ${OpenCV_LIBS})
+add_definitions(-O3 -pthread)
+

yolop/README.md

@@ -0,0 +1,82 @@
+YoloP TensorRT Usage
+=====
+
+Fork from [hustvl/YOLOP](https://github.com/hustvl/YOLOP), modify from [ausk/YOLOP](https://github.com/ausk/YOLOP/), push into [wang-xinyu/tensorrtx](https://github.com/wang-xinyu/tensorrtxls)
+
+
+## 1. Prepare building environments
+
+Make sure you have install `c++`(support c++11)、 `cmake`、`opencv`(4.x)、`cuda`(10.x)、`nvinfer`(7.x).
+
+
+## 2. build yolop
+
+Go to `yolop`.
+
+```
+mkdir build
+cd build
+
+cmake ..
+make
+```
+
+Now you can get `yolop` and `libmyplugins.so`.
+
+
+## 3. Generate and test trt
+
+Go to `yolop/build`.
+
+### 3.1 generate yolop.wts
+Download/Clone [YOLOP](https://github.com/hustvl/YOLOP)
+
+Edit `gen_wts.py` , change `YOLOP_BASE_DIR` to realpath of `YOLOP`.
+
+```
+# [WARN] Please download/clone YOLOP, then set YOLOP_BASE_DIR to the root of YOLOP
+python3 ../gen_wts.py
+```
+
+### 3.2 generate yolop.trt
+```
+./yolop -s yolop.wts  yolop.trt
+```
+
+Now you have such files:  `libmyplugins.so yolop yolop.wts  yolop.trt`
+
+
+### 3.3 test yolop.trt
+```
+mkdir ../results
+
+YOLOP_BASE_DIR=/home/user/jetson/tmp/YOLOP
+./yolop -d yolop.trt  $YOLOP_BASE_DIR/inference/images/
+```
+
+It will output like as follow if successful! ( test on `Jetson Xavier NX - Jetpack 4.4`)
+```
+1601ms # the fist time is slow
+26ms   # then it is faster
+29ms
+27ms
+29ms
+29ms
+```
+
+![](https://user-images.githubusercontent.com/4545060/197756635-38348dc5-d8e7-4ae3-be56-6b231dd2f5db.jpg)
+
+
+
+----------------------------------------
+
+
+```BibTeX
+@misc{2108.11250,
+Author = {Dong Wu and Manwen Liao and Weitian Zhang and Xinggang Wang},
+Title = {YOLOP: You Only Look Once for Panoptic Driving Perception},
+Year = {2021},
+Eprint = {arXiv:2108.11250},
+}
+```
+