Computer Vision Tasks

In this section of documentation, we outline the fundamentals of each computer vision task on which you will build your project. If you are looking for a project creation walkthrough, you will need to head to Tutorials. You may think of the following section as a pre-work, read-through before you start interacting with the user interface.

Once you familiarize yourself with the rudimentary tasks Intel® Geti™ supports, it will become easier to select the right type of task for your problem. This knowledge will save you time and frustration and enable you to select the right type of computer vision task for your problem.

Computer vision tasks you can choose for your project:

• classification guide

• object detection guide

• image segmentation guide

• anomaly classification guide

• task chaining guide

After familiarizing yourself with computer vision tasks, we encourage you to head to the Tutorials section to learn and play around with creating:

• detection project

Note

We strongly encourage you to read the following pages about computer vision tasks in a sequential order. We wrote this section in a topic-based manner. However, computer vision tasks supported by Intel® Geti™ are interrelated. Building your knowledge one task at a time will give you a deeper understanding of Intel® Geti™ capabilities.

It is worth mentioning that we can further divide those tasks into

Supervised learning:

• Classification

• Object Detection

• Image Segmentation

Semi-supervised learning:

• Anomaly Classification

What differentiates supervised learning from semi-supervised learning is the presence of labeled datasets. In supervised learning, you will have to annotate images for an algorithm to learn and increase the accuracy of predicting an outcome. Therefore, you are supervising every aspect of the training data and specifically labelling every aspect you want a model to learn. In semi-supervised learning, you put less effort into annotating images. In semi-supervised methods you typically only partially annotate the data with only one class, the algorithm figures out the rest. However, semi-supervised tasks require a vast amount of computing power and the outcomes are less accurate compared to supervised learning with labeled datasets. On the other hand, supervised learning is time-consuming and may require experts to annotate datasets. Choosing one is clearly a matter of individual approach to your case.