LinkedIn and 3rd parties use essential and non-essential cookies to provide, secure, analyze and improve our Services, and to show you relevant ads (including professional and job ads) on and off LinkedIn. Learn more in our Cookie Policy.

Select Accept to consent or Reject to decline non-essential cookies for this use. You can update your choices at any time in your settings.

Start free trial Sign in

From the course: Safeguarding AI

Image recognition

From the course: Safeguarding AI

Start my 1-month free trial Buy for my team

Image recognition

“

- [Instructor] Image recognition versus object classification, the idea behind both is to find a pattern within an image and identify what that image is. In terms of classification, it will say, "This image contains a dog." In terms of object classification and object detection, this will actually show where that dog appears in the image and draw a boundary box around it. This is usually performed by applying a multi-level neural network on the image, and getting the classification and position information from that network. As mentioned before, people have been researching this area over the last few decades, and have developed a number of extraordinary resources in order to support the creation and training of these machine learning models. As I said, one of the largest sources has been ImageNet, which is a vast tagged archive of images that's used every year in different competitions and is constantly used to measure the state-of-the-art in machine learning. These image troves, in combination with the increase in available computing power, and the increase in research, has led to the gains and improvement in image recognition over the last few decades. This has also led though, to a number of models being supplied for free to the market, including a large one a few years ago called Inception which Google released, which is built off training models from ImageNet over weeks and weeks of computing time. These models can then be downloaded for free and used to enhance current use cases. With all this new technology and new research, cats and dogs aren't the only thing that can be classified from images. A large amount of research has been used in the fashion industry, and it's even got to the point where minute changes in fashion details can be picked up and inferred by these models to help make new design decisions, and create new styles. These kinds of applications have really grown, and it's just amazing to see where the current state-of-the-art is.

Contents