We, humans, are highly visual creatures, and it is clear why. Vision is our primary sense to perceive, understand and interact with everything that surrounds us. It is no wonder why 50% of our brain participates directly or indirectly in visual processing tasks. In fact, neurons dedicated to processing visual information actually take up 30% of our cortex.
This is the reason why, for decades, scientists have tried to provide machines with the ability to see. Artificial Intelligence, which is the field allowing machines to perform human-like tasks, encompasses a great variety of activities, with Computer Vision among them.
In the last few years, we’ve seen companies and organizations adopting different AI solutions to improve processes and efficiency. In fashion eCommerce, for instance, computer vision and AI are being used to classify garments, accessories, and shoes and describe them in a fine-grained manner. Let’s dig a bit deeper into it.
Defining tech buzzwords
As we’ve seen, Artificial Intelligence is any technique that allows computers to perform human-like tasks.
You may have heard of Machine Learning as well. ML is a subset of AI in which machines learn by experience using structured or labeled data fed by humans in order to make informed decisions. An example would be recommendation systems that suggest products to eCommerce visitors based on their website behavior and the products they’ve previously browsed.
Another important term to understand is Deep Learning. DL is a subset within Machine Learning that requires more data but is also able to perform more complex tasks. Deep Learning is characterized by networks made up of layers of artificial “neurons”, inspired by human brain connections. These networks interpret data and are able to identify patterns and relationships. Then, they slightly change and adapt the connections through several stages of data processing to improve the results.
How does AI perform image classification?
Image classification refers to the task of analyzing images in order to categorize them based on different attributes But how does a machine do that?
The process starts with something quite simple —the image pixels. These pixels contain information on color. By analyzing blocks of pixels and the contrast in color between them, AI is able to detect edges, corners, and borders, therefore isolating the important information that will serve to define the features of the objects in the image.
Let’s take an image of a denim shirt, for example. The AI will analyze different groups of pixels in it to detect differences in color. Those differences will help define the limits of the shirt arms and neck. Then, the AI will take into account lines, color and shapes in order to predict that the object in the image has a 90% chance of being a denim shirt.
In deep learning techniques, images pass through several layers of neurons that perform a series of operations before providing an output. After passing through many layers —this is why it is called “deep” learning— the AI is able to output a prediction of the different classes an image belongs to with a certain confidence level. Thanks to these deep layers of neurons—that form what is known as a convolutional neural network— it can predict classes down to very specific details (the color, the pattern, the shape or the length of a dress). From there, it keeps learning and adjusting parameters to provide the most accurate results.
What is its value for fashion eCommerce?
As we explore the full potential of AI applications in the real world, we soon realize it can speed up time-consuming tasks and provide organizations with good value for money.
SaaS AI solutions provide access to advanced technology without large initial investments. In the case of eCommerce, automated product tagging solutions —like the one Catchoom offers— increase operational speed and make products available in less time. This is highly beneficial in terms of revenue. If you want to take a deeper look at that, these infographics will clearly show you the benefits.
Moreover, in cases where fashion retailers lack product information, it generates relevant content with high SEO value in an automated way, therefore powering improved search results that will drive more traffic and boost conversions.