New to the forums? Click here to read the "How To" Guide.

Developer? Click here to go to the Developer Forums.

Facebook’s VR/AR neural network 16x SS on Quest?

Shadowmask72Shadowmask72 Posts: 4,289 Valuable Player
edited September 15 in General

Comments

  • LuciferousLuciferous Posts: 2,550 Valuable Player
    edited September 16
    Impressive and I'm probably being a bit dumb here but how for example does an AI conclude things like the flag should have a diamond in the middle and thin green lines. I understand maybe improving jagged lines and textures.

    Wouldn't it effectively be guessing shapes based on its training which could lead to some very odd visuals?
  • DaftnDirectDaftnDirect Posts: 6,234 Volunteer Moderator
    edited September 16
    I assume it's relying partly on a sequence of frames to build that kind of information, Iike conventional supersamplung relies on moving your head to achieve better resolution, this relies on moving frames, well just my guess.
    Intel 5820K [email protected], Titan X (Maxwell), 32GB Corsair Vengeance DDR4, ASRock X99 Taichi, Samsung 500Gb 960 Evo M.2, Corsair H100i v2 Cooler, Inateck KTU3FR-4P USB 3 card, Windows 10 Pro v2004 (19041.508)
  • LuciferousLuciferous Posts: 2,550 Valuable Player
    edited September 16
    I assume it's relying partly on a sequence of frames to build that kind of information, Iike conventional supersamplung relies on moving your head to achieve better resolution, this relies on moving frames, well just my guess.
    That's what I thought but if the input is not sending different angles to make the shape assumptions, IE I had loaded the game but left my headset on the desk in the same position, the image would be a blurry mess at first? If it relied on moving it must alternate in quality dramatically. Also you may move your head but the angle on the flag may still not reveal its true shape from your perspective. So when you walk up to it may change shape with better information.

    I know not exactly related but I am listening to some LiteRPG books at the moment on Audible. Nearly all of the games worlds in the genre are controlled and created by AI. Looking forward to the day of intelligent NPC's in a AI built Skyrim.
  • saami81saami81 Posts: 293
    Nexus 6
    I believe that AI would be biggest thing in this decade. A life changer. Not just in gaming, but for everything.
  • DaftnDirectDaftnDirect Posts: 6,234 Volunteer Moderator
    edited September 16
    It probably doesn't need more than a couple different frames so would happens real quick, maybe blurry for 2 or 3 frames?... in which case you'd be hard pushed to see it.

    Edit: plus some elements of a frame can be intelligently worked out like larger shapes and geometry... straight edges etc, so I'm guessing even the first frame would look ok.
    Intel 5820K [email protected], Titan X (Maxwell), 32GB Corsair Vengeance DDR4, ASRock X99 Taichi, Samsung 500Gb 960 Evo M.2, Corsair H100i v2 Cooler, Inateck KTU3FR-4P USB 3 card, Windows 10 Pro v2004 (19041.508)
  • snowdogsnowdog Posts: 7,935 Valuable Player
    The crazy fools. This is the start of SkyNet. :o
  • ZenbaneZenbane Posts: 16,411 Valuable Player
    edited September 16
    how for example does an AI conclude things like the flag should have a diamond in the middle and thin green lines

    This type of stuff was done before using a "best guess" technique referred to as "nearest neighbor" logic (algorithm).

    So in your example, would the original "dot" in the middle of a flag "most likely" (nearest neighbor) be a circle, square, bunny, bird, star, etc.

    Nearest Neighbor example:
    https://github.com/vee-upatising/DLSS

    But Nearest Neighbor is most often found within standard Machine Learning techniques, where in the OP we may be going beyond that with Neural Networks; which try to emulate how the Human Brain rationalizes a "best guess."

    So in this case, the Neural Network may simply understand, beforehand, what all the flags in the world look like, and can make a best guess based on that pre-loaded information. The same way we would do that if we saw a blurry image of a flag ourselves.
  • ZenbaneZenbane Posts: 16,411 Valuable Player

    I love this article! It gets extra points for using characters from one of my first, and most favorite, VR Experiences of all time: Farlands
     <3 


Sign In or Register to comment.