Lesson 24 AI Technology on Images
Recently, AI technologies which utilize images have been dramatically developed, and their applications are widely spreading. This chapter describes the outline of the technologies on images.
♦ Image Classification
AI will detect what is captured in the whole area of an image.
[Dog]
[Cat]
♦ Object Detection
AI will create a rectangle in an image, and detect what is captured in the rectangle.
This is also valid in the case of including multiple items in an image.
♦ Semantic Segmentation
AI will categorize each pixel.
An explanation on pixel is available Lesson 15 . As you see below, you can confirm the color of each pixel by magnifying an image.
Semantic segmentation is to get results on each pixel. Therefore, this is suitable for item detection which is difficult to be surrounded by a rectangle. In the following example, the corresponding pixels to a person and a monkey in the image are coloredin blue and green respectively.
Generally, the size of the calculation results increases in the order of image classification, object detection, and semantic segmentation. Therefore, the time for the deep leaning increases in the same order. When you create your own AI, it's important to design it so that you can minimize the size of the calculation result. It will get your deep learning time shorter, and as a result, the time to acquire your own AI for your job will be shorter.