Meta recently decided to jump into the pool of artificial intelligence. For quite a few months, in which other technology companies have bet very heavily on this field, it almost seemed that it was so focused on other fronts that it had completely neglected this one. What’s more, until the end of February, the most transcendental thing in this regard was that Yann LeCun, the company’s head of AI, stated that ChatGPT was not that innovative, a statement that should be somewhat nuanced.
At the end of February, however, released flame (Large Language Model Meta AI), a model (not a service, as is the case with ChatGPT, the new Bing and Bard, among others), that the company makes available to certain sectors through a non-commercial license. This was a big step forward for the company, but still, because its scope is quite limited, it didn’t do much to publicize its AI capabilities.
This is more important than it may seem at first. Let’s remember that the company has had to face quite negative consequences as a result of its commitment to the Metaverse, a proposal as fantastic as it is fanciful, and that just over a year after it was presented in style, it seems condemned to ostracism. The impact that this has had on the company, on its accounts and on its image has been more than considerable, and if we take into account that it was already low for multiple reasons, it is evident that you need, more than ever, to improve your image in any way.
This explains why Meta surprised us today by presenting SAM, an AI model capable of identifying the various elements of an image. Available both for download and execution on our own system, as well as in a trial version through a website designed specifically for this purpose, SAM (Search Anything Model) is capable of analyzing the images, in order to identify each one of them. the elements (people, objects, etc.) that compose them.
For this purpose, the model has been trained with a dataset of 1,000 million images, which has also been shared by Meta, along with quite a bit of technical documentation on the dataset and the model. On the test page it shows a selection of it, with 50,000 images in which we can check how SAM works, but it also allows us to upload our own images in order to check how it works on them.
Detecting elements in images is not something new. For example, we have already talked on some occasion about the magic eraser function of Google Photos. However, that Meta has fully released SAM makes your reach much more global. In addition, functions such as the one that allows you to “cut” a selected element to use it later separately, expand their usefulness exponentially.
