Periodically, we're using this blog to profile some of the companies that participated in the 2009 Emerging Companies Summit. You can learn more about innovative companies that use NVIDIA's GPU technology in the GPU Ventures Zone.
Have you ever watched an entire video, trying to find yourself, only to realize you don't pop up till the last 10 seconds? With video content on the rise, there's a growing need for a fast and efficient way to index video. Text-based tags are a partial solution- but what we really need is true computer vision that can unlock information inside video. Viewdle
, a startup with offices in Los Angeles and Kiev, Ukraine, has come up with a GPU-based solution using facial recognition and visual analysis that lets consumers and enterprises quickly and accurately find and tag people in videos.
At the Emerging Companies Summit
, Viewdle CEO and President Laurent Gil demonstrated
how Viewdle's engine uses sophisticated algorithms to recognize the 'fingerprint' of a person's face. Once Viewdle knows a person's face, that information can be shared on any of three platforms: the cloud, desktop or mobile. With Viewdle, you can search online news feeds for a clip of a particular celebrity or figure; create a slideshow of your family that pulls from your entire video collection; or use your phone to capture video of your friends, automatically recognize who they are before you're even done recording and tag the clip for upload to social networks.
Viewdle started in the Ukraine
with military research into object recognition and computer vision. But commercial opportunities abound. 'We want to be the Adobe of visual analysis,' said Gil in a conversation with NVIDIA VP of Business Development Jeff Herbst. 'License the tool and give the reader away free.' And while there are others in this space, Gil believes Viewdle's methods are superior. Its algorithms have been developed for high performance in uncontrolled environments with various lighting conditions, camera angles and resolutions. The company's visual analysis uses statistical data that looks at the whole face rather than just features.
None of this would be possible without GPUs. As Gil said, Viewdle's engine is 'written from the ground up with parallel processing. You break videos into frames, you find the fingerprints of people and run it all at once.' Not only does the GPU architecture make Viewdle's offerings possible, it does so at a relatively low hardware cost. The examples Gil gave at ECS ' such as analyzing 10 video feeds in real time or processing thousands of facial comparisons per cycle ' were all made on an NVIDIA GTX 9800
Viewdle is already working with photo and video editing software houses to integrate its technology into their products. Able to process 100 million facial comparisons in roughly two minutes, its engine offers consumers the chance to painlessly tag their growing photo and video collections. It expects to develop further touch points between platforms when it releases its Tegra-based
mobile application in early 2010, providing recognition at the point of capture.