Ferhan Özkan’s Post

Transcript

When we originally started Area, we actually built Area just for internal use. But then we realized that it will be so much more impactful to actually give Aria to the broader research community because there are many, many technical challenges within the egocentric machine perception. We believe we need a whole research community to work on these hard problems to be able to make progress. We don't think we can do it alone. The Audio Research Kit, which Meta provides to its approved research partners, is our first attempt to build the future AR glasses with the broader research community. The Aria classes have a rich sense of suite, all encompassed into a variable form factor being about 70 grams. The Machine Detection Services allows researchers to harness algorithms used internally at Meta such that partners can focus their energy on what matters most for their research. Machine perception services are cloud APIs. Researchers can upload the data that they collect using their devices to these machine perception services and they can get outputs like very accurate location trajectories, I gaze vectors from array data and also point clouds of the environment. The tools include a mobile companion app both for Android and iOS which allows you to connect to the classes and control the classes. And then there are very user friendly interface tools on web for desktop that allows you to manage data, access data, process data from the classes. The SDK allows you to stream data in real time from the array classes over to a PC so that you can do prototyping. In addition, we open sourced face and license plate blurring model called ego blur such that the research community can use this model to blur faces and license plates and image data. The research institution started to use Aria for a broad range of research topics including embodied AI, contextualized AI, human computer interaction, robotics, just to name a few. It's just so natural to use that research is really love this tool. We are fully committed to open source. We believe that best minds across the world can use this powerful tool that we've built to help progress the field. On top of the Aria devices, we also offer open source data set. Do a few examples, then reach out to our support groups in all the channels that we have available. We have very passionate researchers and engineers waiting for any questions and to engage with the research community. We encourage the research community to open source the datasets and models that they built using area. The I feel is moving so fast. The time to apply for audio research kit is now.