SIGGRAPH 2018 Papers: Machine Learning, Graphics, and Rendering | Hacker News Reader

SIGGRAPH 2018 Papers: Machine Learning, Graphics, and Rendering

medium.com

92 pointsingve11 comments

No mention of Real Time Live [1]? That was the most exciting session in SIGGRAPH this year! The combination of insane real time rendering and real time capture is going to make for some amazing graphics experiences in the near future.

[1] https://www.youtube.com/watch?v=rPam5CHFQMQ

Here is the full list of papers with links to preprints for the papers presented at SIGGRAPH 2018:

http://kesen.realtimerendering.com/sig2018.html

Here [1] is an overall video overview over some selected papers in this years SIGGRAPH

[1] https://www.youtube.com/watch?v=t952yS8tcg8

And here's [1] Stephen Hill's list of links for 2018:

[1] http://blog.selfshadow.com/2018/08/16/siggraph-2018-links/

FTA : Deep Convolutional Priors for Indoor Scene Synthesis.

Anybody think of a huge improvement to NetHack, DwarFortress and other generated worlds ?

One of the siggraph papers is titled ‘FontCode: Embedding Information in Text Documents using Glyph Perturbation’. This is a type of steganography where the shape of the font is perturbed in a way to encode information. Basically, you make a key value store (a code book in their jargon) of input letters mapped to subtle variations of the same letter of a specific font. My confusion arises from how you get Microsoft word to use these variations. Since the font file on your pc contains only one glyph for each letter, how do you get a text editor to use bastardized glyphs that are not in the OS?

The authors are assuming that they are allowed to use custom fonts. So if they need eight different variants of any particular glyph, they can just create eight different custom fonts where each glyph has the desired appearance.

And yes, if you received a Word document with custom fonts, you might be suspicious of hidden information, but the authors are further assuming that the document is only distributed in a format where the use of custom fonts is not readily apparent. This is why they talk about extracting the information from "an encoded document stored as a vector graphic or pixel image, or even on a printed paper".

Wow. The GAN texture synth is so accurate. And only requires ~2 hours training time on Nvidia Titan GPU (~$1600) ;)

PyTorch code here:

http://vcc.szu.edu.cn/research/2018/TexSyn

AR/VR is getting pretty hot these days. What happened to those old recommendations of sitting 8 ft from a television? Now we put the VR screens 1 inch from our eyes lol.

Modern displays don't use cathode ray tubes, and in VR the optics between the display and the eyes focuses the latter as if the display was a lot more distant that it really is, so these recommendations are less relevant here.

OLEDs are a step above even LCDs in emulating natural light and preventing the glare that makes staring at cathode ray tube monitors less than a good idea.

Reply on news.ycombinator.com