YOLOv7

The Pytorch implementation is WongKinYiu/yolov7.

The tensorrt code is derived from QIANXUNZDL123/tensorrtx-yolov7

Contributors

Requirements

TensorRT 8.0+
OpenCV 3.4.0+

Different versions of yolov7

Currently, we support yolov7 v0.1

For yolov7 v0.1, download .pt from yolov7 release v0.1, then follow how-to-run in current page.

Config

Choose the model tiny/v7/x/d6/w6/e6/e6e from command line arguments.
Check more configs in include/config.h

How to Run, yolov7-tiny as example

generate .wts from pytorch with .pt, or download .wts from model zoo

// download https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-tiny.pt
cp {tensorrtx}/yolov7/gen_wts.py {WongKinYiu}/yolov7
cd {WongKinYiu}/yolov7
python gen_wts.py
// a file 'yolov7.wts' will be generated.

build tensorrtx/yolov7 and run

cd {tensorrtx}/yolov7/
// update kNumClass in config.h if your model is trained on custom dataset
mkdir build
cd build
cp {WongKinYiu}/yolov7/yolov7.wts {tensorrtx}/yolov7/build
cmake ..
make
sudo ./yolov7 -s [.wts] [.engine] [t/v7/x/w6/e6/d6/e6e]  // serialize model to plan file
sudo ./yolov7 -d [.engine] [image folder]  // deserialize and run inference, the images in [image folder] will be processed.
// For example yolov7
sudo ./yolov7 -s yolov7.wts yolov7.engine v7
sudo ./yolov7 -d yolov7.engine ../images

check the images generated, as follows. _zidane.jpg and _bus.jpg
optional, load and run the tensorrt model in python

// install python-tensorrt, pycuda, etc.
// ensure the yolov7.engine and libmyplugins.so have been built
python yolov7_trt.py

INT8 Quantization

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
unzip it in yolov7/build
set the macro USE_INT8 in config.h and make
serialize the model and test

More Information

See the readme in home page.