Author Topic: ray histogramm fusion - is this something?  (Read 3425 times)

2014-06-13, 10:28:10

damjan

  • Active Users
  • **
  • Posts: 13
    • View Profile
i stumbled upon this paper for siggraph 2014:
http://dev.ipol.im/~mdelbra/rhf/  (youtube video included)
just curious to hear some thoughts if this really can work as advertised, because the speedups look huge..


2014-06-13, 11:11:20
Reply #1

racoonart

  • Active Users
  • **
  • Posts: 1446
    • View Profile
    • racoon-artworks
That looks quite impressive O_o
Definitely something I want to hear more about, especially Ondras opinion of course.
Any sufficiently advanced bug is indistinguishable from a feature.

2014-06-13, 11:39:02
Reply #2

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12768
  • Marcin
    • View Profile
I'm not any kind of specialist so what I say might be stupid but isn't this pretty close to rendering hd-cache/light cache (some kind of biased path tracing) with interpolated samples? If yes, then this has to produce some visible bias.
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2014-06-13, 11:51:41
Reply #3

racoonart

  • Active Users
  • **
  • Posts: 1446
    • View Profile
    • racoon-artworks
As far as I understand it it's more like "filter out information that is too far from the average". That is pretty different from merging and interpolating samples from different locations (like irradience map samples). So there's not a problem with artifacts but in some scenes (e.g. some little bright lit spot on a wall which would be hit by GI bounces and illuminate a wall on the other side of the room) it filters out subtle "useful" information - but that also happens with the msi threshold.
Any sufficiently advanced bug is indistinguishable from a feature.

2014-06-13, 15:39:03
Reply #4

racoonart

  • Active Users
  • **
  • Posts: 1446
    • View Profile
    • racoon-artworks
Oh wait, I think I misread this, the description says:
Quote
Each pixel in the image is characterized by the colors of the rays that reach its surface. The proposed filter uses a statistical distance to compare with each other the ray color distributions associated with different pixels, at each scale. Based on this distance, it decides whether two pixels can share their rays or not.

So, that sounds indeed like some kind of interpolation.

Anyways, It would be awesome for animations. Rendering stuff without pre-baking and such reduced render times is such a cost and time saver that it's even worth some little artifacts if there should be some. For me, it sounds like a way to make pathtracing useful for production ;)
Any sufficiently advanced bug is indistinguishable from a feature.

2014-06-13, 17:26:09
Reply #5

borisquezadaa

  • Active Users
  • **
  • Posts: 614
    • View Profile
But how it will behave with small details?... seems suitable for large areas of same color at large resolutions...
What i do with Corona My Corona post of random stuff rendering
WARNING: English.dll still loading...

2014-06-13, 17:45:57
Reply #6

damjan

  • Active Users
  • **
  • Posts: 13
    • View Profile
But how it will behave with small details?... seems suitable for large areas of same color at large resolutions...

check the pdf, there are more examples - leaves, grass volumetric effects etc. it seems to work fine. the real question is how it works with textures

2014-06-13, 19:03:25
Reply #7

arqrenderz

  • Active Users
  • **
  • Posts: 996
  • https://www.behance.net/Arqrenderz1
    • View Profile
    • arqrenderz