Do you want to get deeper into the world of annotation? If so, keep reading this blog!
As a volunteer or a partner, you might have often heard us talk about annotation and how we work with people affected by conflict to upskill them and prepare them for work in the annotation sector.
Annotation is an important part of the tech sector and in this article, we will demystify annotation and give you some examples of the type of annotation methods our beneficiaries use when they start on work or training projects with us!
The definition of image annotation
Image annotation is important in computer science and can be characterized in a variety of ways. The following are two definitions of image annotation:
- Image annotation makes it easier for machines to identify items in images. We help the computer grasp what’s in the image by adding further details. When the model encounters a large amount of similar data, it learns to recognize objects more accurately. The accuracy with which we label our data determines the quality of our outcomes. We must be careful while labeling objects in order to create reliable computer vision models.
- Image annotation is also the process of adding information to images for computer vision models. Keywords are used by robots to automatically label photos.
There are many different types of image annotation; we will look at 2 of them: bounding box and polygon.
Bounding box annotation is a well-known technique for detecting correct objects by drawing rectangles from one corner to the other. We annotate with a rectangle drawing of lines from one corner to another of the object in the image as per its shape to make it fully detectable in bounding box annotation.
A bounding box is an imaginary rectangle that acts as a reference point for object detection and generates a collision box for that object. It can be determined using the x and y axis coordinates in the upper-left corner and the x and y axis coordinates in the lower-right corner of the rectangle. Bounding boxes are widely used in object detection and localization applications. Bounding boxes are one of the most used methods for image and video annotation.
We recognize that bounding boxes are quick and easy to use, however they are inefficient when dealing with irregular forms. When it comes to real-world environments, irregular forms outnumber their regular counterparts by far. Bounding boxes are only available in rectangles and squares. In contrast, polygon annotation captures more lines and angles and can be quite precise depending on how the item appears. In polygon annotation, we click at specific locations to plot polygonal points.
Polygon annotation allows annotators to modify the direction of an object as needed to properly reflect its true shape and size. Polygons are used by annotators to draw lines around the outer edge of the item they want to annotate, defining the shape of the object.
For labeling irregular shapes, polygons are suitable. Annotation that is pixel-perfect indicates that no irrelevant pixels are included inside the annotated area.
Depending on the intricacy of the item, annotating polygons can take much longer than annotating bounding boxes. Not all annotation tools allow you to draw holes inside polygons or indicate that two polygons in a truncated image belong to the same object.
Ghazaleh Alhabib, Platform Coordinator, Humans in the Loop