Computer Vision for Deep Learning – a brief introduction

[suchona.kani.z]

What is Computer Vision (CV)?
Computer vision (CV) is a field of computer science that involves the interpretation of images and videos by machines. Today, deep learning techniques are most commonly used for computer vision. The uses and advantages of using convolutional neural networks (CNNs) are shown below. These are neural network architectures that are modeled on the way the human eye works. The methods presented can also be applied to videos without any problems.

What is Deep Learning?
Deep learning is a sub-discipline of artificial intelligence that is based on the use of deep neural networks. The advantage over normal machine learning is the ability to train end-to-end. This means that you no longer give the deep learning network individual input variables, but only an image or a sentence. The processing of images into features is greece consumer email list carried out automatically by the neural network. At the end, a prediction is made and the neural network can automatically calculate the error and optimize itself based on feedback.

disciplines of computer vision
There are special disciplines in computer vision based on different problems. The simplest discipline is classification: a neural network receives an image and then assigns one or more classes to the image. The AI ​​recognizes what can be seen in the image and outputs a probability for the best-fitting class. The best-known network is probably the ResNet-50.

In the following image, you can see a zebra and below it are the probabilities of what type of object it is. The model says that there is a 99.5% probability that it is a zebra. However, the other suggestions are still shown for the sake of completeness.

image recognition of a zebra
Semantic Segmentation
Semantic segmentation involves classification at the pixel level. It is possible to create images in which each pixel is assigned to a class. A well-known network for this purpose is the U-Net. Segmentation is particularly important for precise demarcation and exact localization of objects.
Examples of this are tumor segmentation and autonomous driving. The zebra can be seen again in the following image, this time all pixels belonging to the zebra are colored red, while those of the grass and the sky have other colors.