Signed Distance Field rendering - pros and cons (as used in PS4 title Dreams) *spawn

Shifty Geezer said:
They actually go over a number of ways of using the data which'll save others the R&D. Some of the results looked fabulous but weren't suitable for their specific UGC based game. The final solution uses 2D splats based on a point cloud evaluation.

Also, three years experimenting! It's all very well hoping other devs will try alternative rendering techniques, but few can afford the luxury of three+ years experimenting without a working product to sell!
Click to expand...

Yes it is only possible for AA title with a publisher. It is not as costly as a AAA title but they are 35 to work on the title. And AAA title can't do this and I think it is too risky for indie.
 
Last edited:
Shifty Geezer said:
Also, three years experimenting! It's all very well hoping other devs will try alternative rendering techniques, but few can afford the luxury of three+ years experimenting without a working product to sell!
Click to expand...
I felt like I was talking about a last gen renderer after hearing Alex's speak just before my turn. Geometry aliasing and triangle mesh LOD are real problems once you can render nearly million objects per frame (most of them being small and far away). At least our pipeline partially sidesteps the quad efficiency problem, but the other problems are impossible to solve as triangles cannot be prefiltered. I hate triangles as much as Alex, but unfortunately we have to live with them for some more years.

Triangles up close + SDF based LOD is an interesting idea. I expect to see hybrid solutions late this gen. Unreal already supports SDF based shadows and AO.
 
For SDF based representation does/will the art pipeline stay the same or will they need/want to modify the way things are done? I am looking through media molecules slides but haven't gotten very far yet. In the meantime can someone brief me on how animation (ala skinned meshes) are handled?
 
Skinned meshes don't exist. MM's solution doesn't even include textures. The base modelling is an assemblage of CSGs. The monochrome models in the slides are representative. Each CSG can have a colour and these are blended. The CSG's can also have other properties defined, so some are dense and resolved in detail, while others are sparse and resolved as larger splats. Colour and texture appears to be created using 2D art techniques, so varying the qualities of the splats.
 
Shifty Geezer said:
Skinned meshes don't exist. MM's solution doesn't even include textures. The base modelling is an assemblage of CSGs. The monochrome models in the slides are representative. Each CSG can have a colour and these are blended. The CSG's can also have other properties defined, so some are dense and resolved in detail, while others are sparse and resolved as larger splats. Colour and texture appears to be created using 2D art techniques, so varying the qualities of the splats.
Click to expand...
I know there no skinned meshes, I meant a method/technique with similar functionality. There does seem to be animated models so I was wondering how they handled that. All I could infer is that they might be transforming the base primitives and doing something other than union and subtraction between them.

Oh and for those interested in signed distance fields I currently perusing the following:
edit - rendering worlds with two triangles
http://www.iquilezles.org/www/material/nvscene2008/rwwtt.pdf
and
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.480.140&rep=rep1&type=pdf

edit - also this page is pretty nice: http://iquilezles.org/www/articles/distfunctions/distfunctions.htm
 
Last edited:
Yes, that seems to be the case, moving the underlying CSGs. Their 'union' support soft blends, so an arm made of objects should join at the shoulder in a convincing way with blended colours/properties.
 
You must log in or register to reply here.
Back
Top