Fb Outlines Advances in Laptop Imaginative and prescient and Object Identification Tech
While machine learning systems are much better at identifying objects in still images, the next stage in that process is identifying individual objects in videos, which could open up new considerations on brand placement, visual effects, accessibility features, and much more.
Google has been developing its tools in this area for a while now, which has now brought about new advances in YouTube’s options, including the ability to tag Products displayed in video clips offer direct shopping options, enabling wider e-commerce opportunities in the app.
And now Facebook is also taking the next steps with a new process with which individual objects in video images can be singled out much better.
As explained by Facebook:
“Working with researchers from Inria, we developed a new method called DINO that allows Vision Transformers (ViT) to be trained unsupervised. In addition to defining a new state of the art among self-supervised methods, this approach leads to a remarkable method result that is unique for this combination of AI techniques. Our model can detect and segment objects in an image or video without any surveillance and without a segmentation target. ”
This effectively automates the process, which is a great advance in computer vision technology.
And as mentioned earlier, this opens up a number of new potential opportunities.
“Segmenting objects makes things easier, from swapping the background of a video chat to teaching robots to navigate a crowded environment. This is seen as one of the toughest challenges in machine vision as the AI really needs to understand what is contained in an image. Traditionally, this is done with supervised learning and requires a large number of annotated examples. However, our work with DINO shows that high-precision segmentation may only be solvable with self-supervised learning and a suitable architecture. “
This could help Facebook provide new options like YouTube for labeling products for associated display in video content. As Facebook notes, there are also applications related to AR and visual tools that could lead to much more advanced and comprehensive Facebook features.
This could also include further data collection and personalization.
Back in 2017, in the early stages of its video recognition efforts, Facebook realized that advances in technology would lead to increased capacity to present users with more relevant content based on their viewing habits.
“AI inference could evaluate video streams, personalize the streams for individual users’ newsfeeds, and reduce latency in publishing and distributing videos. Personalizing real-time reality video could be very compelling and, in turn, increase the time users spend in the Facebook app. “”
Of course, Facebook probably wouldn’t be as overt now in its goals when it comes to getting users to consume more time-consuming content – but that’s of course the goal of bringing the most compelling and valuable experience to all users. to maximize exposure time and increase its utility and value.
This also offers more advertising opportunities – and again, it’s easy to see how important these advanced video recognition tools can be to Facebook’s advertising business. In the YouTube example, the plan is to tag all items in all video clips, not just those that the creator assigns a tag to provide more shippable product options across the app.
Whether or not YouTube takes this step we will have to wait and see, but it is interesting to see what other implications such advances will have and how they can transform your marketing and advertising process.
And then there is AR. With Facebook developing its own AR glasses, it is also possible that this technology could be used to better identify objects in your real worldview and provide support, promotions and other information.
There are a multitude of potential use cases, and it’s interesting to see how Facebook’s tools develop on that front.
The complete DINO research paper and the findings can be found here.