Image annotation tool

Semantic segmentation

This is the most commonly used form of image segmentation. It involves separating each pixel in an image into classes and then labeling them.Annotators use a polygon annotation tool to outline objects so that they can accurately capture their shape. Any background features, like sky or trees, are also assigned labels and highlighted. This means that every pixel in the image is connected to a label.

Instance segmentation

The annotation type adds further granularity to images. Whereas semantic segmentation assigns every occurrence of an object to the same class, instance segmentation identifies each time an object occurs. When there are more than one of the same object in an image then both will be given their own label and highlight colour. By doing this it is possible to further enrich image training data.

Bounding box

This is the most common annotation type. Using the Keylabs image annotation tool annotators drag boxes around targeted objects. This technique is quick and simple. However, it does not capture as much detail as other methods.

Polygon

The Keylabs platform allows annotators to outline complex shapes. By connecting small lines around an object annotators can precisely define its shape. This process is known as polygon annotation.

Rotating bounding box

In the image data annotation process, the rotating bounding boxes tool provides a way to more accurately encapsulate objects that aren't aligned with the horizontal or vertical axes of the image. By allowing the bounding box to rotate around its center, the annotation can closely match the orientation of the object, ensuring more precision in datasets, especially for tasks like object detection where the object's orientation matters. This approach minimizes the unnecessary background within the bounding box tool, enhancing the accuracy of machine learning models trained on such annotated data.

Skeletal

Many AI applications are designed to interpret human bodies. Skeletal annotation trains AI models to do this. Annotation tools allow annotators to attach lines to images of humans, locating limb positions and body shapes.

Cuboid

Cuboid data annotation tools enable the representation of objects in images with three-dimensional properties, capturing depth, width and height within a two-dimensional space. These data annotation tools place 3D bounding boxes, or cuboids, around objects, providing a richer spatial understanding compared to traditional 2D bounding boxes. Primarily employed in autonomous driving and augmented reality applications, cuboid annotations give computer vision models a nuanced understanding of object placement and scale within a scene.

Key points annotation

This technique uses points to locate important features. One common example of key point annotation is facial feature labeling, where points are used to identify noses, eyes, lips, etc.

Mesh

Mesh data annotation tools are instrumental in creating intricate, polygonal representations of objects within images or 3D spaces. Unlike traditional annotation methods, these tools plot vertices and edges to form a dense mesh, encapsulating the nuanced contours and surfaces of an object. Widely used in 3D modeling and advanced computer vision tasks, mesh annotations capture the complex geometric and topological characteristics of objects.

Lane annotation

This annotation technique is primarily used to identify roads, railway lines, pipelines and other linear structures. Annotators trace the shape of these features using annotation platform tools.

3D point cloud

A 3D point cloud is considered to be a more complex type of annotation, because it is already about 3D. From some types of cameras, information comes in three-dimensional scanning of space and not in the form of videos and pictures. Basically, we are now talking about Lidar - this is a radar, a device that receives a three-dimensional description of objects. This is a slightly more complex annotation level.

Bitmask

This annotation type links individual pixels to specific objects, allowing for annotations that are separated or contain gaps.

Custom annotation

All of the above annotation techniques can be combined to accomplish specific annotation goals. Keylabs image annotation platform online gives AI companies flexibility and options when planning annotation projects.