duan8/yolov3-tiny

yolov3-tiny

The Pytorch implementation is ultralytics/yolov3 archive branch.

Excute:

1. generate yolov3-tiny.wts from pytorch implementation with yolov3-tiny.cfg and yolov3-tiny.weights, or download .wts from model zoo

git clone -b archive https://github.com/ultralytics/yolov3.git
// download its weights 'yolov3-tiny.pt' or 'yolov3-tiny.weights'
// put tensorrtx/yolov3-tiny/gen_wts.py into ultralytics/yolov3 and run
python gen_wts.py yolov3-tiny.weights
// a file 'yolov3-tiny.wts' will be generated.

2. put yolov3-tiny.wts into tensorrtx/yolov3-tiny, build and run

// go to tensorrtx/yolov3-tiny
mkdir build
cd build
cmake ..
make
sudo ./yolov3-tiny -s             // serialize model to plan file i.e. 'yolov3-tiny.engine'
sudo ./yolov3-tiny -d  ../../yolov3-spp/samples // deserialize plan file and run inference, the images in samples will be processed.

3. check the images generated, as follows. _zidane.jpg and _bus.jpg

Config

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

More Information

See the readme in home page.