Skip to content

Latest commit

 

History

History
125 lines (92 loc) · 4.53 KB

README.md

File metadata and controls

125 lines (92 loc) · 4.53 KB
Raw

yolov8

The Pytorch implementation is ultralytics/yolov8.

The tensorrt code is derived from xiaocao-tian/yolov8_tensorrt

Contributors

Requirements

  • TensorRT 8.0+
  • OpenCV 3.4.0+

Different versions of yolov8

Currently, we support yolov8

Config

  • Choose the model n/s/m/l/x from command line arguments.
  • Check more configs in include/config.h

How to Run, yolov8n as example

  1. generate .wts from pytorch with .pt, or download .wts from model zoo
// download https://github.com/ultralytics/assets/releases/yolov8n.pt
cp {tensorrtx}/yolov8/gen_wts.py {ultralytics}/ultralytics
cd {ultralytics}/ultralytics
python gen_wts.py -w yolov8n.pt -o yolov8n.wts -t detect
// a file 'yolov8n.wts' will be generated.
  1. build tensorrtx/yolov8 and run

Detection

cd {tensorrtx}/yolov8/
// update kNumClass in config.h if your model is trained on custom dataset
mkdir build
cd build
cp {ultralytics}/ultralytics/yolov8.wts {tensorrtx}/yolov8/build
cmake ..
make
sudo ./yolov8_det -s [.wts] [.engine] [n/s/m/l/x]  // serialize model to plan file
sudo ./yolov8_det -d [.engine] [image folder]  [c/g] // deserialize and run inference, the images in [image folder] will be processed.
// For example yolov8
sudo ./yolov8_det -s yolov8n.wts yolov8.engine n
sudo ./yolov8_det -d yolov8n.engine ../images c //cpu postprocess
sudo ./yolov8_det -d yolov8n.engine ../images g //gpu postprocess

Instance Segmentation

# Build and serialize TensorRT engine
./yolov8_seg -s yolov8s-seg.wts yolov8s-seg.engine s

# Download the labels file
wget -O coco.txt https://raw.githubusercontent.com/amikelive/coco-labels/master/coco-labels-2014_2017.txt

# Run inference with labels file
./yolov8_seg -d yolov8s-seg.engine ../images c coco.txt

Classification

cd {tensorrtx}/yolov8/
// Download inference images
wget  https://github.com/lindsayshuo/infer_pic/blob/main/1709970363.6990473rescls.jpg
mkdir samples
cp -r  1709970363.6990473rescls.jpg samples
// Download ImageNet labels
wget https://github.com/joannzhang00/ImageNet-dataset-classes-labels/blob/main/imagenet_classes.txt

// update kClsNumClass in config.h if your model is trained on custom dataset
mkdir build
cd build
cp {ultralytics}/ultralytics/yolov8n-cls.wts {tensorrtx}/yolov8/build
cmake ..
make
sudo ./yolov8_cls -s [.wts] [.engine] [n/s/m/l/x]  // serialize model to plan file
sudo ./yolov8_cls -d [.engine] [image folder]  // deserialize and run inference, the images in [image folder] will be processed.

// For example yolov8n
sudo ./yolov8_cls -s yolov8n-cls.wts yolov8-cls.engine n
sudo ./yolov8_cls -d yolov8n-cls.engine ../samples
  1. optional, load and run the tensorrt model in python
// install python-tensorrt, pycuda, etc.
// ensure the yolov8n.engine and libmyplugins.so have been built
python yolov8_det.py  # Detection
python yolov8_seg.py  # Segmentation
python yolov8_cls.py  # Classification

INT8 Quantization

  1. Prepare calibration images, you can randomly select 1000s images from your train set. For coco, you can also download my calibration images coco_calib from GoogleDrive or BaiduPan pwd: a9wh

  2. unzip it in yolov8/build

  3. set the macro USE_INT8 in config.h and make

  4. serialize the model and test

More Information

See the readme in home page.