Separate names with a comma.
Discussion in 'HardForum Tech News' started by HardOCP News, Aug 11, 2014.
You guys really have to check out this video of Microsoft's new Hyperlapse video technology.
That is awesome.
Very, very cool; would love to watch some hiking videos or tours of areas using that.
This is pretty neat. I like the idea behind it especially when doing tour videos of something. I'm sure it will find many great uses for demos.
i wonder how this works
well ... that was cool ..
If I would take a guess.. by eliminating frames with excess angle change.. as to how one goes about it, i don't know.. I suppose they studies and averaged what constitutes 'jerkiness' in a video (say maybe it is 4-frames per fast angle change) then you can compare somehow frames for excess change in content or something, you know like first four vs the next four then vs the next four, and then if there is too much change they vote out or something... I don't know I am making stuff up for fun
Interesting stuff. My guess is they somehow merge 10 frames either before and/or after into a single frame and do so for each frame depending on the amount of motion within each frame.
At any rate.. I'm waiting for Apple to announce this new technology next year.
Along with the new apple ipro camera.. hehe.. however they do it.. it is quite mesmerizing.
Incoming line of hipsters, snarkily bemoaning how MS stole this from Apple.
Cool I dig it
Same here and Super awesome
Looks like a video game with the textures that pop in, especially with the mountain climbing parts.
that is completely awesome!
wow that really cool. Well done
When I want to make videos like that I just ride my bike really fast.
That's pretty neat
RTFA, tech details inside, 35mb pdf, more vids. .
It is cool in theory, but a mind-numbingly boring video seen 10 times faster is still a mind-numbingly boring video.
wow this is pretty bad ass.
Maybe I could finally convert the many, many hours of GoPro SCUBA videos I have.
True, but if you take a semi-interesting video and make it 10 times faster, you end up with a 5 times interesting video
Nope. You can read the research paper. Short version is that the scene is reconstructed in 3d and a virtual camera path is constructed, then the scene rendered from video frame data. Think about something like the photosynth technology applied to video.
This explains most of it. Totally worth the watch.
Looks like you are playing a fast paced FPS game from back at the turn of the century.
I dont recall graphics looking like that at the turn of the century, what games where you playing ? and what were you smoking ?
Yes, I read the article (well, I went to the website and watched the technical 'how they did it'.)
Long story short: They create a 3D recreation of your path based on your camera inputs, then plot a new smooth path through your footage. They then render a new video, based on data from multiple frames that contain the same view, but as if it came from the stabilized smooth path.
Obviously it's not perfect yet, but it is damned impressive already. (During the mountain climb part, you can see his hand "melt" in and "melt" out of view, and his co-hiker half-disappear before completely disappearing.) But still, damned impressive. I hate to imagine how much computing power this takes, though. First they have to analyze the footage to create a 3D-space reconstruction of your route, then computationally determine a smooth path that has complete data sets. The final render is the 'easy' part.
And of course, as soon as I post, I see there was a page 2 of comments that already said what I posted.
I'm convinced that MS actually has a bunch of hamsters running the show, while the actual employees are secretly spending all of their time elsewhere. How is this anything amazing?! Speed it up and smooth it? C'mon now.
What are you smoking that would make you think the graphics were what I was referring to?
I take it you didn't watch the technical video. It's actually really fucking impressive.
If i'm reading the article right. The camera doesn't actually exist???
They basically split the video into individual frames, extrapolated the geography based on the images, and then spliced the images together to generate a new perspective where the original camera isn't located.
I've seen this done by bolting a 360 degree laser scanner on top of a car to generate the point cloud data. But they got this out of a freaking gopro!
Did they say what kind of hardware did the rendering and how long it took?
I looked at it briefly. I didn't find it that intriguing. I have no doubt that it was tedious labor.
Given the rock face changes and the motion tweenesque effect, i would assume they are inferring what the scene should look like in 3d, similar to photosynth tech.
pretty cool, the artifacting on the climbing videos was interesting