Pages

Facebook is training Artificial Intelligence to help it recognize objects using 1 billion Instagram images, which does not need human-labeled images to learn accurately 'see'

Instagram is a platform where people share and post over a billion photos on a daily basis which include sceneries, mountains, pugs, Ferraris, brunches, and babies. Most people use this app for sharing photos because these photos are the memories of all of us. Instagram has become the number 1 image database all over the world over the last decade. Now the owner company of Instagram is trying to teach machines what are the importance of photos in our life and what photos show us, how these present the memories, and how what is inside the photos. Therefore, Facebook researchers are training a self-supervised algorithm using over a billion images on Instagram provided that this algorithm will not need human-labeled images to recognize what is inside the image.

Because the accurate image recognition algorithm needs human to tag photos as containing an animal in it like a dog, horse, cow, etc., only then this algorithm works and recognize the resemblances between different images in which human has specified, having the same subject. The scientist of Facebook Yann LeCun has been given this duty to change A.I reliance on labels for many years. This computer vision program is nicknamed SEER which gave an accuracy result of 84.2% result when it was put into test by ImageNet which is a big visual database made for use in visual object recognition software research. Facebook tested it to check whether it recognize clearly what object is shown in the image. Facebook said that it has trained SEER to identify the different objects in an image by analyzing random, not labeled images of Instagram.

Further, this A.I will be trained in such a way in the near future that it will recognize directly from the information that is given, whether it is given in the form of text, photos, or any other sort of information, without depending on human-labeled images. The good performance of SEER shows that it can surpass the computer vision tasks in the real world, still, this is a research project.

During the research, they include better automatic generated text for elaborating the images to people who are visually impaired, and to make automatic categories of items that are sold on the Facebook marketplace, and keep the harmful images blocked on the Facebook platform, the company said in a blog post.

But many users may not like one thing that their images are used to train A.I, but they need not worry about because their images are used only for research and innovation purpose, because it is included in the policy of Instagram and Facebook that they can use the data for research purpose.

Sources: 1 / 2.

Read next: Facebook is rolling out a feature in over 90 regions which will allow users to turn off political content from appearing onto their feeds

No comments:

Post a Comment