Press "Enter" to skip to content

Facebook Uses 1 Billion Public Photos to Train AI

Instagram is filled with pictures and it is one of the largest image databases on the planet. It is the treasure of images. The Facebook researchers have found a new AI model that can learn from Random pictures that are available on the internet.

They added a million publicly available images without labels or annotations. The computer vision program is called SEER. This achieved an accuracy score of 84.2%. It is tested whether it can identify what is in the picture. SEER learned how to identify objects in photos by analyzing random, unlabeled, and uncurated Instagram images. This AI technique is known as self-supervised learning.

Many Instagram users were surprised to hear this information. They were worried about their privacy.

Priya Goyal, a software engineer at Facebook AI Research said that Facebook has mentioned to its users that it can access the data of the users for research and innovation. The company also said that it will open-source the software so that the researcher can work with and understand the depth of it. Other big tech companies like Google and Microsoft are also working hard to push the boundaries of computer vision. Google published the SimCLRv2 computer vision model, while OpenAI published iGPT 2.

The research team developed an all-purpose PyTorch-based library for self-supervised learning called VISSL, which is open-sourcing to encourage the broader AI community to test with the technology. This will allow them to create more flexible, accurate, and adaptable computer vision. They will also automatically create annotations for visually impaired users. The company also said the users to keep them stashed away from harmful images.