Facebook debuts new video datasets for training neural networks
Facebook Inc. unveiled its latest contribution to the artificial intelligence ecosystem at the GitHub Universe conference on Thursday.
Researchers from the social network have put together two datasets designed to help with creation of machine learning models that process video content. Typically, the information used in AI projects doesn’t receive as much attention as the sophisticated technologies that underpin the development process. But it’s an equally important component that in many cases can be harder to obtain.
Before they’re ready to launch, AI models must be trained against sample data to hone their accuracy and performance. It’s often the case that the more advanced the task a neural network is built to perform, the more fine-grained the training information has to be. The issue is that the readily available datasets out there don’t cover every use case.
As a result, companies often have to manually assemble and organize information for their projects. That can be a major time sink given that upwards of gigabytes worth of data may be required. Facebook’s two new datasets aim to speed up the task for projects that implement computer vision, one of the most widely used forms of AI.
The first dataset bears the descriptive name of Scenes, Objects, and Actions. It’s made up of videos that have been manually tagged with information about their contents. That includes the setting of each clip, the objects featured in the footage and what they’re doing.
Manohar Paluri, the head of Facebook’s computer vision team, said that the file is geared toward training fairly advanced models. He used the example of a hypothetical neural network designed to identify kayaks that can not only detect when they appear in a video but also infer if they’re in use.
The second dataset that Facebook unveiled, which is called General Motions, serves a more straightforward purpose. It’s a collection of GIFs that each display a specific physical motion such as jumping or sliding.
There’s a wide set of use cases where Facebook’s datasets could be applied. Computer vision technology can be found in self-driving cars, drones, a growing number of smart home appliances and multimedia management, among other services. There are so many applications that Intel Corp. paid a hefty $400 million for a startup called Nervana Systems Inc. last year to obtain technology that puts it in a better position to capitalize on this market.
A message from John Furrier, co-founder of SiliconANGLE:
Your vote of support is important to us and it helps us keep the content FREE.
One click below supports our mission to provide free, deep, and relevant content.
Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.