Meta represents an AI that can identify an object it has never seen before. An innovation aimed in particular at showing that the giant is a major player in the sector.
Artificial intelligence (AI) usually needs to be trained on existing content in order to be able to detect objects, but Meta has found a way that this technology can do this unaided. The social media giant has released a “Segment Anything”artificial intelligence model that can detect objects in images and videos, even if they were never part of its training data. You can select items by clicking on them or by using text prompts. As Reuters explains, you can type in the word “cat”and watch the AI highlight all the felines in any photo.
Meta introduces an AI that can identify an object that has never been seen before
This model can also be paired with others. This can help you reconstruct an object in 3D using a single image or draw from mixed reality headset representations. Segment anything to limit the need for additional training.
The AI model and dataset will be available for download with a license for non-commercial use. In other words, creators cannot use it for their products. An initiative that primarily focuses on research and increased access to technology. Meta currently uses a similar technology for content moderation, post recommendation, and photo tagging.
The developers acknowledge that the existing model is biased. This one can miss finer details, and it’s not as good at edge detection as some models. And while Segment Anything can process hints in real time, it quickly shows its limitations when working with large images. Some of the more specialized AI tools should perform better in their field, Meta says.
An innovation aimed in particular at showing that the giant is a major player in the sector.
You shouldn’t see this AI in robots or other devices where fast and accurate object detection is vital. At the same time, such models should still be able to help in situations where relying solely on training data is impossible or hardly possible. A social network can use technology to stay afloat in the face of a rapidly growing volume of content. And this at least shows that Meta aims to generalize computer vision.
Meta is no stranger to sharing its achievements in the field of artificial intelligence, such as its unwritten language translator. That being said, the company is always under pressure to show that it can weigh as much in this market as tech giants like Google and Microsoft. It’s already working on implementing generative AI “characters”for its social apps, and innovations like Segment Anything show it has some very good tricks up its sleeve.