YOLO Series TensorRT Python/C++

简体中文

Support

YOLOv8、YOLOv7、YOLOv6、 YOLOX、 YOLOV5、YOLOv3

• YOLOv8
• YOLOv7
• YOLOv6
• YOLOX
• YOLOv5
• YOLOv3

Update

• 2023.8.15 Support cuda-python
• 2023.5.12 Update
• 2023.1.7 support YOLOv8
• 2022.11.29 fix some bug thanks @JiaPai12138
• 2022.8.13 rename reop、 public new version、 C++ for end2end
• 2022.8.11 nms plugin support ==> Now you can set --end2end flag while use export.py get a engine file
• 2022.7.8 support YOLOv7
• 2022.7.3 support TRT int8 post-training quantization

Prepare TRT Env

Install via Python

pip install --upgrade setuptools pip --user
pip install nvidia-pyindex
pip install --upgrade nvidia-tensorrt
pip install pycuda
pip install cuda-python

Install via C++

By Docker

Try YOLOv8

Install && Download Weights

pip install ultralytics

Export ONNX

from ultralytics import YOLO

model = YOLO("yolov8s.pt")
model.fuse()  
model.info(verbose=False)  # Print model information
model.export(format='onnx')  # TODO: 

Generate TRT File

python export.py -o yolov8n.onnx -e yolov8n.trt --end2end --v8 --fp32

Inference

python trt.py -e yolov8n.trt  -i src/1.jpg -o yolov8n-1.jpg --end2end 

Python Demo

Expand

1. YOLOv5
2. YOLOx
3. YOLOv6
4. YOLOv7

YOLOv5

!git clone https://github.com/ultralytics/yolov5.git

!wget https://github.com/ultralytics/yolov5/releases/download/v6.1/yolov5n.pt

!python yolov5/export.py --weights yolov5n.pt --include onnx --simplify --inplace 

include NMS Plugin

!python export.py -o yolov5n.onnx -e yolov5n.trt --end2end

!python trt.py -e yolov5n.trt  -i src/1.jpg -o yolov5n-1.jpg --end2end 

exclude NMS Plugin

!python export.py -o yolov5n.onnx -e yolov5n.trt 

!python trt.py -e yolov5n.trt  -i src/1.jpg -o yolov5n-1.jpg 

YOLOX

!git clone https://github.com/Megvii-BaseDetection/YOLOX.git

!wget https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_s.pth

!cd YOLOX && pip3 install -v -e . --user

!cd YOLOX && python tools/export_onnx.py --output-name ../yolox_s.onnx -n yolox-s -c ../yolox_s.pth --decode_in_inference

include NMS Plugin

!python export.py -o yolox_s.onnx -e yolox_s.trt --end2end

!python trt.py -e yolox_s.trt  -i src/1.jpg -o yolox-1.jpg --end2end 

exclude NMS Plugin

!python export.py -o yolox_s.onnx -e yolox_s.trt 

!python trt.py -e yolox_s.trt  -i src/1.jpg -o yolox-1.jpg 

YOLOv6

!wget https://github.com/meituan/YOLOv6/releases/download/0.1.0/yolov6s.onnx

include NMS Plugin

!python export.py -o yolov6s.onnx -e yolov6s.trt --end2end

!python trt.py -e yolov6s.trt  -i src/1.jpg -o yolov6s-1.jpg --end2end

exclude NMS Plugin

!python export.py -o yolov6s.onnx -e yolov6s.trt 

!python trt.py -e yolov6s.trt  -i src/1.jpg -o yolov6s-1.jpg 

YOLOv7

!git clone https://github.com/WongKinYiu/yolov7.git

!wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-tiny.pt

!pip install -r yolov7/requirements.txt

!python yolov7/export.py --weights yolov7-tiny.pt --grid  --simplify

include NMS Plugin

!python export.py -o yolov7-tiny.onnx -e yolov7-tiny.trt --end2end

!python trt.py -e yolov7-tiny.trt  -i src/1.jpg -o yolov7-tiny-1.jpg --end2end

exclude NMS Plugin

!python export.py -o yolov7-tiny.onnx -e yolov7-tiny-norm.trt

!python trt.py -e yolov7-tiny-norm.trt  -i src/1.jpg -o yolov7-tiny-norm-1.jpg

C++ Demo

support NMS plugin show in C++ Demo

Citing

If you use this repo in your publication, please cite it by using the following BibTeX entry.

@Misc{yolotrt2022,
  author =       {Jian Lin},
  title =        {YOLOTRT: tensorrt for yolo series},
  howpublished = {\url{[https://github.com/Linaom1214/TensorRT-For-YOLO-Series]}},
  year =         {2022}
}

Sponsor

Buy me a cup of coffee