Supported tasks

With the Intel® Geti™ platform you can perform multiple types of tasks that will train a model for your own use case. The tasks stay in the realm of computer vision, which means you will be annotating images or video frames.

In the following section, we will give you an overview for each type of task, provide links for detailed explanations of these tasks, and links to video tutorials that take you step by step through the process of creating an AI model with the Intel® Geti™ user interface.

Supported tasks
	Classification is characterized by assigning a label to an image or video frame. You define labels upon starting your project and assign one of them to the image from your dataset. The computer then learns how to classify the whole image or video frame. To learn more about the classification task, read our guide.
	Detection is characterized by rectangular or square shapes that encompass objects in an image or video. You draw that shape around the object and give it a name or as we call it in computer vision - a label or class. The computer then can track that object in a video or identify it in an image. To learn more about the object detection task, read our guide or watch our tutorial.
	Image segmentation is characterized by delineated shapes surrounding objects in an image or video. You draw polygons contouring the shape of the object. The computer then highlights the isolated region and groups every pixel in that delineated shape for later processing. To learn more about an image segmentation task, read our guide or watch our tutorial.
	Anomaly project is characterized by the absence of annotating a dataset by a human. You provide a set of images or videos with regular patterns. Subsequently, you provide a set of images with anomalous patterns. The computer then learns to distinguish between normal and anomalous. To learn more about an anomaly project, read our guide or watch our tutorial.

Hint

To learn how to train a model that will let you automate the tasks described above, walk through our tutorials.

Task Chaining

The task chaining project is characterized by combining previously mentioned types of projects into one single flow. You choose a pre-defined task chain and proceed with the rules of annotation for classification, detection, or segmentation. To learn more about task chaining, read our guide.

Pre-defined templates for supported tasks

We can further divide the supported tasks into pre-defined templates for each supported task.

Classification
	Classification single label is characterized by choosing a single label per image.
	Classification multi label is characterized by choosing multiple labels per image.
	Classification hierarchical is characterized by choosing a single label and multiple labels per image in a hierarchical label structure.

Detection
	Detection bounding box is characterized by drawing a rectangle around an object in an image.
	Classification multi label is characterized by drawing and enclosing an object within a minimal rectangle.

Segmentation
	Instance segmentation is characterized by detecting and delineating each distinct object of interest in an image.
	Semantic segmentation is characterized by grouping parts of an image that belong to the same object.

Anomaly
	Anomaly classification is characterized by choosing and categorizing images as normal or anomalous.
	Anomaly detection is characterized by detecting and categorizing objects as normal or anomalous.
	Anomaly segmentation is characterized by segmenting and categorizing objects as normal or anomalous.

Chained tasks
Detection > Classification	Detection > Classification is characterized by localizing objects and drawing rectangles around them followed by categorizing detected objects.
Detection > Segmentation	Detection > Segmentation is characterized by delineating objects’ shapes followed by categorizing segmented objects.