Ambient Occlusion

[Wumpf]

Some kind of realtime Ambient Occlusion. Most likely we will end up using HBAO (screen space). However, a more global technique non-screen-dependet technique would be really cool (but also probably too compilicated for our limitated resources)

[Wumpf]

Depends on #7 and #51

[Wumpf]

Notes to myself:

• There are ridiculously many papers to screen space ambient occlusion/obscurance.
• HBAO is certainly NOT the holy grail of SSAO. It's just another try to make the original implementation better... while being terrible slow
• These nice guys (nvidia again, but MUCH newer 2012 - HBAO is 2008) do not only introduces just another one, but also have a small overview in their slides
• Looks like most impls need normals. But since we have mostly flat surfaces, we can probably get them really easy from depth derivatives.
• Cubes. We have only cubes. There MUST be room for simplifications!

[Wumpf]

• THIS is an excellent overview
• And a-simple-and-practical-approach-to-ssao is truly simple and practical :). If I'am not wrong it also resembles some ideas of Alchemy AO
• This is similar to the gamedev article, but different in some ways
• My basic idea by now:
  • Do weighting of samples like in alchemy
  • Use random samples in a screenspace circle (since it avoids unnecessary oversampling). Sadly - as the GamDev-Article points out - I have to scale this one with the inverse depth, since otherwise feeling of perspective would be gone.
  • New: (in this combination, haven't it seen in any other impl) Force the sample to be within the right half-space. Don't now yet how to do that, but this should be possible and useful somehow. Need to investigate.
• Doesn't look like there is any possibility to optimize for OUR cube-scenario other than tweaking

[Wumpf]

We have now a working version of Alchemy SSAO. Only High-Quality setting so far. Need lower quality fallbacks. aeb7e7c72c291a88be56e82095844ac2f81634d9

[Wumpf]

Lower quality fallbacks were introduced in 1dcbfbb72a4a3363e53a072434dbf8e64a91b638. Was stable since then. However a close look into #62 is needed.