yolov7

The Pytorch implementation is WongKinYiu/yolov7.

environment

win10 vs2019 cuda10.2 cudnn7.6.5 tensorrt7.0

Different versions of yolov7

Currently, we support yolov7 v1.0

• For yolov7 v1.0, download .pt from yolov7 release v1.0, git clone -b v1.0 https://github.com/WongKinYiu/yolov7 and https://github.com/QIANXUNZDL123/tensorrtx-yolov7.git, then follow how-to-run in current page.

Config

• Choose the model tiny/v7/x/d6/w6/e6/e6e from command line arguments.
• Input shape defined in yololayer.h
• Number of classes defined in yololayer.h, DO NOT FORGET TO ADAPT THIS, If using your own model
• INT8/FP16/FP32 can be selected by the macro in yolov5.cpp, INT8 need more steps, pls follow How to Run first and then go the INT8 Quantization below
• GPU id can be selected by the macro in yolov5T.cpp
• NMS thresh in yolov5T.cpp
• BBox confidence thresh in yolov5T.cpp
• Batch size in yolov5T.cpp

How to Run, yolov7-tiny as example

1. 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_trts.py {WongKinYiu}/yolov7
cd {WongKinYiu}/yolov7
python gen_wts_trtx.py -w yolov7-tiny.pt -o yolov7-tiny.wts
// a file 'yolov7-tiny.wts' will be generated.

2. to change your cuda path and opencv path also tensorrt path

cd {tensorrtx-yolov7}/
// update CLASS_NUM in yololayer.h if your model is trained on custom dataset
change model_check = "-s" -s is generate engine   also model_check = "-d" is run engine