YOLO5+TENSORRT acceleration

1.Precautions before use

If you are using the YAHBOOM version of the image, there is no need to build the environment. If you are building the environment yourself, you need to download the Tensorrt package The version of yoloV5 should correspond to the package of Tensorrt This tutorial uses the YOLO5 V5.0 version, so the Tensorrt acceleration package also needs to use V5.0 This is the Tensorrt download link：https://github.com/wang-xinyu/tensorrtx/tree/master/yolov5

2.start using

1. Gen under tensorrt/yolov5 Copy gen_wts.py to the folder of yolov5
2. Execute gen_wts.py build .wts file。

python3 gen_wts.py yolov5s.pt

3. Go to the yolov5 folder under the directory tensorrtx as usual, create a build file, and enter

xxxxxxxxxx
mkdir build
cd build
cmake ..

4. Classify in yololayer. h_ Modify NUM to yours. The official dataset is Coco, so the default is 80. (This step can be ignored if it is official)
5. 执行makeFile。(Make once every time it is modified to ClassNUM)

xxxxxxxxxx
make -j4

6. Copy the wts file from the file path of yoloV5 to tensorrtx/yolov5.
7. build .engine file

xxxxxxxxxx
    sudo ./yolov5 -s ../yolov5s.wts yolov5s.engine s

3.Can testing achieve acceleration

way1：Enter the folder tensorrtx/yolo5/build and run the command

xxxxxxxxxx
sudo ./yolov5 -d yolov5s.engine ../samples

Can be run in the YOLOv5 folder

xxxxxxxxxx
python3 detect.py

By comparison, it can be clearly seen that the addition of Tensorrtx greatly reduces the recognition time of images

way2：running

xxxxxxxxxx
cd tensorrtx/yolo5
python3 yolov5_trt.py

It can also be run in the YOLOv5 folder

xxxxxxxxxx
python3 detect.py

By comparison, it can be clearly seen that the addition of Tensorrtx greatly reduces the recognition time of images