rendering training

[jeromeetienne]

transparency

• blending
  • blending srcA, oneMinusSrcA
  • color and alpha
• first render, opaque object
• second render transaparent face
  • back to front
  • depth buffer read only
  • so sorted every rendering
• http://www.opengl.org/wiki/Transparency_Sorting

[jeromeetienne]

mipmap

• suppose you got a big texture that is displayed small on screen
• so the filtering is taking only a very limited of pixel
• so it is quite possible that from one frame to another, all the pixels read in the texture are fully differents
• as a consequences, there is a flickering
• mipmap is made to fix this issue

[jeromeetienne]

margin in atlas

• put 1 pixel top/bottom/left/right on EACH tile
• fill this border with bleeding
• so a original tile of 512x512 will be 510x510 usefull
  • uv must be expressed in those 510x510
• if mipmap... in theory, the margin has to be double at each mipmap level
  • gasp... not doable in practice

[jeromeetienne]

storage for him

• simple binary format
• stored in webgl buffer format
• no parsing time

[jeromeetienne]

prototype of paging to load the world

• big block, more district like, than small 1m3 one
• load all block in background
  • store in cache possible
  • local storage, appcache, google apps cache
• no need for actual data
  • so easy to experiment

[jeromeetienne]

ligthing

• he advices more lightmaps than global illumination
  • ok if model + light === static
• http://en.wikipedia.org/wiki/Image-based_lighting
• http://ict.usc.edu/pubs/Image-Based%20Lighting.pdf
• http://www.pauldebevec.com/IBL2003/

analysis

• better than global illumination
• more textures
• but artists driven

[jeromeetienne]

some links

• cube map generation http://code.google.com/p/cubemapgen/
• mipmap http://www.arcsynthesis.org/gltut/Texturing/Tut15%20Needs%20More%20Pictures.html