Meta Launches a New AI Tool That Can Identify Images

Ever wanted to Separate Items in Pictures? well now you can with Segment Anything!

Meta, what with the current mess the metaverse in, Meta (Facebook) has decided to go all in with AI. One of the new additions to their tool kit is “Segment Anything“. This is an AI image identification tool.

Segment Anything’s main aim is to help users identify items in an image with just a few clicks. Please note this is still in demo mode while meta work out the kinks. Meta are boldly stating that this AI tool can already take a photo and individually identify the pixels within the image through segmenting the image.

“Segmentation—identifying which image pixels belong to an object—is a core task in computer vision and is used in a broad array of applications, from analyzing scientific imagery to editing photos,”
Meta wrote in a post announcing the new model

Meta says that the Segment Anything AI system has been trained over a whopping 11 million images, which is of course a large scale of data go through. Another noteworthy topic is that Meta is making Segment Anything available for the research community. This will be under a permissive open license, that can be accessed through the Segment Anything Github page. So if you are interested on having a nosey at this project why not go have a look.

Here is an example of a picture being used by Segment anything. It clearly selects parts of the image and breaks them up in this case it has selected the dog on the right.

Meta Moves Away From The NFT Community and Focuses More on AI

All this comes as a shift from meta away from the NFT community. Only just back In February, Meta had started to focus on other products, including artificial intelligence. It has announced that is is focusing on generative A.I. The company has reportedly laid off over 10,000 workers after the end of it’s NFT project on Instagram.

Generative artificial intelligence referes to an A.I. system that generates text, images, or other media in response to a set of prompts. Stable diffusion, chatGPT and Midjourney are all good examples of this use of technology.

The Main Concern is Of Course Again Privacy

One of the key aspects of running these types of generative AI is the large ammounts of data the collect and store. With Meta’s Segment Everything it has used and stored 11 million images. Now the question is was the data collected with prior permission. For example if we use AI for facial recognition then we should be asking permission from individuals to use their face as part of the data set. ChatGPT is already in Hot water over privacy laws across europe.

One positive note to take from this though is that Meta have various privacy-preserving techniques, in a battle to protect user privacy. Blurring faces and other personally identifying information as well as having the option for users to report offensive content. This can be done through their contact email: segment-anything@meta.com with the id of the image. They will then remove it from the dataset if it has proven to be invalid. Which does bring another question does this mean that these picture were taken randomly without obtaining permission?

Conclusion

Here at So Artificial we are quite intrigued by this piece of new tech. The possibility of it’s use cases are quite extensive, but we can’t help that it is just a ploy to be used for surveillance which might have a more sinister meaning. Unless Meta are working on a new art tool that has a lot of AI features. We do like the fact that’s it’s open source and there is an option to report bad images. But we do also feel some of these images may have been collected without prior permission.

We will be keeping a close eye on segment anything as it continues to be worked on over the coming months/years. What do you lot think of this new AI tool?