Skip to content

Latest commit

 

History

History
 
 

yolov3

README.md

yolov3

The Pytorch implementation is ultralytics/yolov3 archive branch. It provides two trained weights of yolov3, yolov3.weights and yolov3.pt

This branch is using tensorrt7 API, there is also a yolov3 implementation using tensorrt4 API, go to branch trt4/yolov3, which is using ayooshkathuria/pytorch-yolo-v3.

Config

  • Input shape defined in yololayer.h
  • Number of classes defined in yololayer.h
  • INT8/FP16/FP32 can be selected by the macro in yolov3.cpp
  • GPU id can be selected by the macro in yolov3.cpp
  • NMS thresh in yolov3.cpp
  • BBox confidence thresh in yolov3.cpp

How to run

  1. generate yolov3.wts from pytorch implementation with yolov3.cfg and yolov3.weights, or download .wts from model zoo
git clone https://github.com/wang-xinyu/tensorrtx.git
git clone -b archive https://github.com/ultralytics/yolov3.git
// download its weights 'yolov3.pt' or 'yolov3.weights'
cp {tensorrtx}/yolov3/gen_wts.py {ultralytics/yolov3/}
cd {ultralytics/yolov3/}
python gen_wts.py yolov3.weights
// a file 'yolov3.wts' will be generated.
// the master branch of yolov3 should work, if not, you can checkout cf7a4d31d37788023a9186a1a143a2dab0275ead
  1. put yolov3.wts into tensorrtx/yolov3, build and run
mv yolov3.wts {tensorrtx}/yolov3/
cd {tensorrtx}/yolov3
mkdir build
cd build
cmake ..
make
sudo ./yolov3 -s                          // serialize model to plan file i.e. 'yolov3.engine'
sudo ./yolov3 -d ../../yolov3-spp/samples // deserialize plan file and run inference, the images in samples will be processed.
  1. check the images generated, as follows. _zidane.jpg and _bus.jpg

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 yolov3/build

  3. set the macro USE_INT8 in yolov3.cpp and make

  4. serialize the model and test

More Information

See the readme in home page.