mark

Annotation is an indispensable step in the production of custom data sets. How to carry out the annotation conveniently and efficiently is the point we need to pay attention to. This paper shares two open source annotation tools, which are also the most used and efficient graphical annotation tools at present. They are LabelImg and CVAT.

LabelImg

LabelImg is a graphical tagging tool written in Python and Qt, supporting Windows, Linux and macOS. LabelImg is a graphical tagging tool written in Python and Qt, supporting Windows, Linux and macOS.

The LabelImg operation is handy because it provides a default class that you can remove if you don’t need those types

Next, you can Open the exe file, click Open to import the image, press the shortcut key W, select the target, the input box will pop up, write the class name, it is ok, if there are more than one target, then continue to mark

LabelImg also supports importing folders. After marking a folder, select Next Image on the left to switch to the Next folder to continue. In the output format part, labelImg currently supports YOLO and PascalOVC2 formats. The former label information is stored in TXT files, while the latter is stored in XML

After finishing the label, you can save the image and label file separately, but the file name is the corresponding, but the extension is different

Finally, look at the contents of the tag file

A line represents a target in the format

class x_center y_center width height
Copy the code

The first column is the index to class, counting from 0, for example, “basketball” and “face”. The last four columns are X_center /image_width, y_center/image_height, width/image_width, height/image_height, and the value ranges from 0 to 1

CVAT

CVAT stands for Computer Vision Annotation Tool. It is an open source Tool developed by Intel. It is very simple to use and can help you quickly create your own data set.

CVAT also provides a website that you can visit at cvat.org/

After registering, create a task

For a simple example, here are some photos of masks to test

Advanced Settings, generally no special requirements, can not set

Once created, you can see it under the Tasks TAB

Once you’re inside, you can mark it

After annotating a picture, click save in the upper left corner, and then click the right arrow above the picture to annotate the next picture until all the pictures are processed

Finally, come to Tasks and export the data set. Currently, CVAT supports COCO 1.0, PASCAL VOC 1.1, MOT 1.1, ImageNet 1.0, YOLO 1.1, etc

A data set in PASCAL VOC 1.1 format is derived

If you need to set up local CVAT, it is also possible, refer to the official set up document github.com/openvinotoo…