jwliu-cc / Label_image_with_medianflow

This is a program that help data makers to label image semi-automatically. It will create and save bounding-boxes and labels in a '.xml' document which is the same as that labeled by 'labelimg.exe'.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Label_image_with_medianflow

This is a program that help data makers to label image semi-automatically. It will create and save bounding-boxes and labels in a '.xml' document which is the same as that labeled by 'labelimg.exe'. It is based on medianflow-traker.

说明:

本程序针对整段视频中所有label都出现且目标运动变化不大的条件

  • lxml,opencv3.4,os,math

    os和math为python原装

    lxml和opencv都可以用conda命令安装

  • 使用方法:

    1. 更改程序中的global parameter以适配自己的环境,注意label数组的内容和顺序。

    2. 运行程序,会在视频所在位置生成储存数据用的文件夹,按R键可以对视频进行旋转,按空格键暂停标注数据,用鼠标框选bounding-box,框选时,绿色框表示框选中,红色框表示框选结束可以添加。按A键将目标添加至跟踪器,注意框选顺序要和label数组中的顺序相同(每框选一次结束后想添加到跟踪器都需要按A确认添加)。添加结束后,再按空格键,视频继续播放,并显示跟踪框选效果,同时开始储存数据。可按S中途停止储存,空格键继续储存。

这是一个demo程序,仍然存在很多bug,操作上有时误按A和空格都会造成数据的错误,因此标注结束后应当进行审查,用labelimg进行审查,更正错误标框。

我的测试结果是:一个10分钟的视频,标注加审核修改的时间用时约30分钟,因跳帧(frame_skip变量)储存,共获得数据图片1111张。

About

This is a program that help data makers to label image semi-automatically. It will create and save bounding-boxes and labels in a '.xml' document which is the same as that labeled by 'labelimg.exe'.


Languages

Language:Python 100.0%