/pinto_trt_yolov7

Primary LanguagePython

TRT YOLOV7

Table of contents

  • Introduction
  • Prerequires
  • How to use
  • Reference

INTRODUCTION

With the original yolov7, after exporting to onnx and finally tensorrt, there is currently a bug at the confidence score, giving the wrong confidence score of the object. Thanks to pinto's yolov7 customization, thus correcting the incorrect confidence score

PREREQUIRES

  • Ubuntu 18.04
  • Tensorrt 8.5.2.2

HOW TO USE

  1. Download yolov7 256x320 model from PINTO model zoo
  2. Use trt exec to convert from onnx model to tensorrt engine
/usr/src/tensorrt/bin/trtexec --onnx=src_model.onnx --saveEngine=dst_model.trt --fp16 --explicitBatch --minShapes=images:1x3x256x320 --maxShapes=images:4x3x256x320 --optShapes=images:2x3x256x320 --workspace=2048
  1. Create "models" folder, then place the model to the models folder + add image folder to run.py. Then,
python3 run.py

REFERENCE:

307_YOLOv7