The One, Two, Threes of Data Labeling for Computer Vision
Although data labeling is also essential for other types of machine learning applications such as natural language processing, speech recognition, and more, for this article we will focus on data labeling for Computer Vision. Specifically:
(1) What is data labeling for computer vision?
(2) Picture examples of data labeling for autonomous vehicles and bird watching
(3) Different options to get data labeling done!
(4) Links to data labeling sources
(1) What is data labeling for computer vision?
Data labeling for computer vision is when we draw a digital outline around objects in a picture so that the computer can distinguish the different parts of the picture for classification to provide a basis for data processing and machine learning.
(2) Picture examples of data labeling for autonomous vehicles and bird watching
Let’s say we want to create a machine learning program for autonomous vehicles. We would need to label the picture so the computer vision of the autonomous vehicle can distinguish the cars from the rest of the picture. To give an example, see the picture below, the cars are labeled with yellow boxes. Now the computer vision can identify the objects in the yellow boxes as cars. This type of labeling is called “bounding boxes”.
Another type of computer vision data labeling is call “polygons”. This type of data labeling is called “polygons” because the boarder of the data labeling is created with multiple continuous line segments that is not a rectangle. See below for the before and after pictures of using the “polygons” type of data labeling to enable our computer vision machine learning program to recognize a bird from the rest of the picture.
Before using “polygons” data labeling, the picture looks like this.
After using “polygons” data labeling, the picture looks like this.
“Bounding boxes” and “polygons” are just two types of computer vision data labeling methods. Other data labeling methods are called, “cuboids, points and lines, semantic segmentation, text annotation, video annotation, and more”. You can read more about them in the links at the end of this article.
(3) Different options to get data labeling done!
In the bird data labeling example above, I was able to do this in seconds, with the help of a computer vision data labeling tool called labelbox.com. You can try it for free to get a quick feel of data labeling.
However, I could see that this would become a tedious task if I were to data label hundreds of bird pictures. That’s why I looked up ways to save my time on data labeling such as hiring a team to do it for you, outsourcing (freelancers or companies that have teams that specialize in data labeling), crowdsourcing (platforms that access large numbers of workers at once), and even machine learning assisted data labeling. You can read more about the different types of data labeling options in the links at the end of this article.
Thanks for reading!
I hope this acticle helped some of you get a quick overview of computer vision data labeling. What I summarized above is just a small part of the data labeling needed to run the machine learning process. What type of data labeling do you use for computer vision? Or in your opinion, what is the most useful tool or method to efficiently do data labeling for computer vision? Leave a note below to start a discussion! Let’s learn through collaboration!
(4) Links to data labeling sources
To learn more about data labeling for machine learning, see the sources below:
https://www.cloudfactory.com/data-labeling-guide
https://labelyourdata.com/articles/what-is-data-labeling-in-machine-learning/
