Scanner: Processing Terabytes of Video on Hundreds of Machines | Hacker News Reader

Scanner: Processing Terabytes of Video on Hundreds of Machines

cs.cmu.edu

79 pointsapoms11 comments

I feel like CMU has been on a tear on HN lately. It feels like there's an incredible amount of momentum around them, from Robotics to Computer Vision to AI/ML. It feels like each thing that comes out lately is even more impressive than the last. Exciting times. Disclaimer: I have no affiliation with CMU whatsoever, just an admirer.

CMU does a ton of neat stuff, but that shouldn't be surprising, they're one of the top CS research universities in the world.

So is ETH but they don't get so much exposure. What is the difference?

HN is primarily US oriented.

HN is primarily Silicon Valley oriented. Pittsburgh is already quite far from the center of gravity.

Scanner is a great tool I have been following it's development very closely over the last year. Great to see it reach 1.0 . I highly recommend checking out Hwang [1] (a sparse video decoder) which is bundled and used in Scanner.

Scanner is one of the first few tools to leverage Docker/Kubernetes by demonstrating ability to ship complex heterogenous architecture in a reliable/reproducible manner.

[1] https://github.com/scanner-research/hwang

I am not an expert in video encoding. Just curious, what does Hwang offer that ffmpeg doesn't?

> Hwang is a library for performing fast decode of frames from h.264 encoded video (most mp4s). Hwang provides both a Python and C++ API. Hwang decodes on the CPU (using ffmpeg) or on the GPU (using the NVIDIA hardware decoder).

Hwang lets you perform "efficient random access" across the video by building a GOP/Segment aware index. This comes in handy in a lot of applications where you want to access particular frame or set of frames but do not wish to decode and store all frames. Most tools such as ffmpeg (command line application not library) are optimized for sequential decode use case.

Oculus Research just rebranded itself as FB Reality Lab. And the Oculus Go stand-alone headset looks like a winner with 100K sales already, 2M+ estimated by years end.

Distributing VR / AR content seems the next big challenge to me. VRChat can handle on order of 10 per room. But we are talking serious grid compute / bandwidth beyond that. Especially for "multiplayer" narrative 360 video ;)

Can't wait until I can `apt-get` install this tool! (I'm not a huge fan of using docker)

Reply on news.ycombinator.com