Skip to content

Latest commit

 

History

History
34 lines (24 loc) · 952 Bytes

README.md

File metadata and controls

34 lines (24 loc) · 952 Bytes
Raw

libtorchYoloV3

First of all, thanks for https://github.com/walktree walktree's great Job.

This repo was modified based on the walktree's job, i've rewritten it with Libtorch v1.6.0 stable version(the lastest one 13th August 2020)

A Libtorch implementation of the YOLO v3 object detection algorithm, written with pure C++. It's fast, easy to be integrated to your production, and CPU and GPU are both supported. Enjoy ~

Requirements

  1. LibTorch v1.6.0
  2. Cpu
  3. OpenCV (4.3.0)

IDE tool Visual studio 2017

Running the detector

The first thing you need to do is to get the weights file for v3:

cd models
wget https://pjreddie.com/media/files/yolov3.weights

load weights\models and images:

change the image path in your main()function.
as 
./{your test image path}/imgs/person.jpg
and weights. cfg as well

As I tested, it will take 2700 ms on Cpu,please run inference job more than once, and calculate the average cost.