Video data annotation tools and platforms (data annotation company)

Video data annotation platform （ Mark outsourcing company ）

The work of data annotation companies is diverse , However, video annotation requires a little higher tools , There are not many platforms that can do it online , Mainly voice 、 Image Overlays . At present, the industry is mixed , Some platforms have strong technical strength , Brand background , For example, JD Zhongzhi 、 Baidu crowdsourcing , Good data confidentiality . Some platforms are specialized as agents , Give him your data needs , He subcontracted it to the next floor . Here are some platforms , It also integrates some opinions of other bloggers , as follows ：

All wisdom in Jingdong

The labeling quality is relatively high , Project delivery on time , The data isolation scheme can be marked without its own server , Pay more attention to customer data security . It also provides privatization Deployment Services .

Baidu public test

The tagging ability is extensive , Baidu has been in the label industry for a long time , Accumulated more crowdsourcing users . But I am not optimistic about crowdsourcing , Because the quality is difficult to control .

figure-eight

Foreign well-known data annotation platform , Many foreign companies have cooperated with it . The demander can configure the annotation tools and corresponding label, Send tasks directly on the platform , No account manager communication … This may not be very friendly to domestic customers .

Video data annotation tool

CDVA

CDVA（compact descriptor for video analysis）, Mainly based on CDVS Compact visual descriptors for video analysis , Previously, compact visual descriptors were mainly used in the field of image retrieval . New datasets need to be created , Annotate video frames , Therefore, according to a blogger's annotation tool on the Internet, some modifications have been made , The function is to select the area to be marked with the mouse in each frame 4 A little bit , The sequence is clockwise . Because the quadrilateral has a wider range , Some people before marked the rectangle directly , But in some affine transformations , Often the positioning effect of rectangle is not good , Rectangle location should be more suitable for face location and pedestrian location . http://www.cnblogs.com/louyihang-loves-baiyan/p/4457462.html

Video annotation tool

Because the laboratory needs to do CDVA Standards for ,CDVA（compact descriptor for video analysis）, Mainly based on CDVS Compact visual descriptors for video analysis , Previously, compact visual descriptors were mainly used in the field of image retrieval . New datasets need to be created , Annotate video frames , Therefore, according to a blogger's annotation tool on the Internet, some modifications have been made , The function is to select the area to be marked with the mouse in each frame 4 A little bit , The sequence is clockwise . Because the quadrilateral has a wider range , Some people before marked the rectangle directly , But in some affine transformations , Often the positioning effect of rectangle is not good , Rectangle location should be more suitable for face location and pedestrian location .

These codes are based on openCV Of , Therefore, it is necessary for the project configuration opencv Library path and header file path . Here is a brief introduction to the usage of this tool

• First run the application to see a black box and a Video window
• Select the area to be marked , Clockwise , Draw 4 A little bit （ When writing this tool, the default is to draw one image per frame , Do not consider multiple situations , If you want to label multiple , You can add an array to access the area of each quadrilateral
• If you need to make further correction if you click the wrong place when drawing, press 'z' that will do , Press down ‘z’ You can go back a point , If finished 4 I still feel dissatisfied , You can also press 'c' Remove all
• When the rectangular area is determined , Press down ‘n’, These data will be written to the specified txt In file , At the same time, it will enter the next frame
• Because the change of continuous frames in the video is not good , Especially when the camera stops , therefore , To avoid drawing the target area repeatedly , The quadrilateral coordinates of the previous frame will be automatically drawn in the next frame , If you need to redraw , Press down ‘c’, that will do , If you don't need to redraw , Press down 'n' The file is written , At the same time, continue to the next frame , So back and forth

In writing txt In file , A row represents the data in a frame , The first number is the number of frames , after 4 Number , They are entered in turn when drawing rectangles 4 Coordinates . You can do it according to your needs , Modify this code , I hope it can help you .

/********************************************************************
    created:	2015/04/18
    created:	18:4:2015   17:24
    filename: 	D:\WorkSpace\VS_Projects\VideoLabel\VideoLabel_Quadrilateral\video_label_quadrilateral.cpp
    file path:	D:\WorkSpace\VS_Projects\VideoLabel\VideoLabel_Quadrilateral
    file base:	video_label_quadrilateral
    file ext:	cpp
    author:		Yihang Lou
    purpose:	draw the quadrilateral labels in the frame captured from video
*********************************************************************/
#include "opencv2/opencv.hpp"
#include <iostream>
#include <string>
#include <vector>
#include <fstream>
using namespace std;
using namespace cv;

// Global variables

Mat img_original, img_drawing;
Point quad [4];
//the value of pointNum is between 0~4
static int pointNum = 0;
/*************************************************
// Method: help
// Description: describe the usage
// Author: Yihang Lou
// Date: 2015/04/18
// Returns: void
// History:
*************************************************/
static void help()
{
    cout << "This program designed for labeling video \n"
            "Only if you press the 'n' the present quadrilateral data will be written into txt file\n";


        cout << "Hot keys: \n"
        "\tESC - quit the program\n"
        "\tn - next frame of the video\n"
        "\tz - undo the last label point \n"
        "\tc - clear all the labels\n"
        << endl;
}
/*************************************************
// Method: drawQuadri
// Description: 
// Author: Yihang Lou 
// Date: 2015/04/18
// Returns: void
// Parameter: quad the point of Point array
// History:
*************************************************/
static void drawQuadri (Point * quad) {
    for(int i = 0; i < 4; i++)
    {
        line(img_drawing,quad[i],quad[(i+1)%4],Scalar(0,255,0),1,8,0);
    }
}


/*************************************************
// Method: onMouse
// Description: do the actions after onMouse event is called
// Author: Yihang Lou 
// Date: 2015/04/18
// Returns: void
// Parameter: event
// Parameter: x Mouse's coordinate
// Parameter: y 
// History:
*************************************************/
static void onMouse(int event, int x, int y, int, void*)
{


    switch (event)
    {
    case CV_EVENT_LBUTTONDOWN:
        quad[pointNum%4].x = x;
        quad[pointNum%4].y = y;
        cout<<"x = "<<x<<" y = "<<y<<endl;
        pointNum++;

        break;
    case CV_EVENT_LBUTTONUP:
        //finish drawing the rect (use color green for finish)

        circle(img_drawing,cvPoint(x,y),1,Scalar(0, 255, 0),1,8,0);

        if(pointNum == 4)
        {
            pointNum = 0;

            cout<<"draw quadri line"<<endl;
            drawQuadri(quad);
        }

        break;
    }

    imshow("Video", img_drawing);
    return;
}

/*************************************************
// Method: isempty
// Description: check the quad is empty
// Author: Yihang Lou 
// Date: 2015/04/18
// Returns: int
// Parameter: quad
// History:
*************************************************/
int isempty(Point * quad)
{
    for (int i = 0 ; i < 4; i++)
    {
        if (quad[i].x !=0 || quad[i].y !=0 )
        {
            return 0;
        }
    }
    return 1;
}
int main(){
    namedWindow("Video");
    ofstream outfile("1.txt");
    help();
    VideoCapture capture("1.avi");
    capture >> img_original;
    img_original.copyTo(img_drawing);
    imshow("Video", img_original);
    setMouseCallback("Video", onMouse, 0);
    int frame_counter = 0;
    while (1){
        int c = waitKey(0);
        if ((c & 255) == 27)
        {
            cout << "Exiting ...\n";
            break;
        }
        switch ((char)c)
        {
        case 'n':
            //read the next frame
            ++frame_counter;
            capture >> img_original;
            if (img_original.empty()){
                cout << "\nVideo Finished!" << endl;
                return 0;
            }

            img_original.copyTo(img_drawing);

            if (!isempty(quad))
            {
                drawQuadri(quad);

                outfile << frame_counter << " " << quad[0].x << " "<< quad[0].y << " " 
                    << quad[1].x << " "<< quad[1].y << " "
                    << quad[2].x << " "<< quad[2].y << " "
                    << quad[3].x << " "<< quad[3].y << " "<<endl;
            }

            break;
        case 'z':
            //undo the latest labeling point
            if(pointNum == 0)
            {
                cout<<"if you want to clear the existent quad please press 'c'"<<endl;
                break;
            }
            pointNum--;
            quad[pointNum].x=0;
            quad[pointNum].y=0;
            img_original.copyTo(img_drawing);
            for(int i = 0 ; i < pointNum; i++)		
            {
                circle(img_drawing,quad[i],1,Scalar(0, 255, 0),1,8,0);
            }

            break;
        case 'c':
            //clear quad array
            memset(quad,0,4*sizeof(Point));
            img_original.copyTo(img_drawing);
        }
        imshow("Video", img_drawing);
    }

    return 0;
}

Copy

VoTT

Visual images released by Microsoft / Video tagging tool . Ability to tag and annotate image catalogs or stand-alone videos . Use Camshift Tracking algorithm assists computer to mark and track objects in video . Export labels and resources to Custom Vision Service CNTK,Tensorflow（PascalVOC） or YOLO Format , For training object detection model . https://github.com/Microsoft/VoTT

vatic

Video annotation tool (vatic)

Github:https://github.com/cvondrick/vatic

1. install （ be based on Ubuntu16.04）

$ sudo pip install cython==0.20
$ wget http://mit.edu/vondrick/vatic/vatic-install.sh
$ chmod +x vatic-install.sh
$ ./vatic-install.sh
$ cd vatic

Copy

vatic-install.sh You may not be able to download ,vatic-install download

2. To configure HTTP Serve

/etc/apache2/sites-enabled/000-default.conf

Copy

Replace with ：

WSGIDaemonProcess www-data python-eggs=/home/cmcross/.python-eggs
WSGIProcessGroup www-data

<VirtualHost *:80>
    ServerName 0.0.0.0
    DocumentRoot /home/cmcross/vatic/public

    WSGIScriptAlias /server /home/cmcross/vatic/server.py
    CustomLog /var/log/apache2/access.log combined
</VirtualHost>

Copy

Enable mod_headers modular ：

$ sudo cp /etc/apache2/mods-available/headers.load /etc/apache2/mods-enabled

Copy

 restart Apache:
$ sudo apache2ctl graceful

Copy

3. To configure SQL

establish vatic database ：

$ mysql -u root
mysql> create database vatic;

Copy

If you need a password to log in , The password is hail_ukraine, modify root The user has no password , Otherwise, an error will be reported , If it's mine above csdn Download the , The password is root

to update mysql root No password

update user set authentication_string=PASSWORD("") where User='root';
update user set plugin="mysql_native_password"; 
flush privileges; # Update all operation permissions 
quit;

Copy

start-up ：

$ cp config.py-example config.py

Copy

If online service is required , modify access Corresponding options , Offline services can skip

Initialize database ：

$ turkic setup --database

Copy

Restart the database ：

$ turkic setup --database --reset

Copy

allow vatic visit turkic:

$ turkic setup --public-symlink

Copy

4. Verify that the installation is correct

$ turkic status --verify

Copy

If you receive any error messages , It means that the installation is not completed .

Be careful ： I'm not going to use Mechanical Turk, Ignore by Mechanical Turk Any errors caused .

ERROR:

Localhost: http://localhost/

Testing access to Amazon Mechanical Turk... ERROR! Signature or access key missing
Testing access to database server... OK
Testing access to web server... ERROR! HTTP Error 403: Forbidden

One or more tests FAILED!

Copy

solve ： modify Apache To configure /etc/apache2/apache2.conf add to

<Directory /home/cmcross/vatic/public>
         Options Indexes FollowSymLinks
         AllowOverride All
         Require all granted
</Directory>

Copy

 Solve the problem after restarting $ sudo apache2ctl graceful

Copy

5. Example

Video framing

$ mkdir /path/to/output/directory
$ turkic extract /path/to/video.mp4 /path/to/output/directory

Copy

The width and height attributes are ：--width 1000 --height 1000 perhaps --no-resize

$ turkic extract /path/to/video.mp4 /path/to/output/directory --width 1000 --height 1000

Copy

The video frames that have been taken can be converted into vatic The format of

$ turkic formatframes /path/to/frames/ /path/to/output/directory

Copy

Import video ( offline )

$ turkic load identifier /path/to/output/directory Label1 ~Attr1A ~Attr1B
 Label2 ~Attr2A ~Attr2B ~Attr2C Label3 --offline

Copy

identifier Is the indicator ,Label1 Will have properties Attr1A and Attr1B,Label2 Will have properties Attr2B,Attr2B and Attr2C, also Label3 Will not have attribute . Specifying attributes is optional

pulish video ( offline )

$ turkic publish --offline

Copy

ERROR：publish Open after http://localhost?id=1&hitId=offline The URL shows Server Error

resolvent ： stay /etc/apache2/sites-enabled/000-default.conf add to

<Directory /path/to/vatic>
    <Files server.py>
        Require all granted
    </Files>
</Directory>

Copy

export voc Format datasets

$ turkic dump identifier -o /output --pascal --pascal-skip 1

Copy

–pascal-skip: How many frames to get data once , Without this attribute, the default is 15 Frame take once

For more parameter references Github:https://github.com/cvondrick/vatic