Landscape/terrain feature request Bicubic Interpolation

The terrains in UE4 would be greatly improved if Bicubic interpolation was implemented. The net effect of this would mean a 2017 terrain would look better than a 4033 terrain and the blocky artifacts from normal terrain would be pretty much gone.

You might ask why I know this and I can say that I worked at Bongfish as a technical artist for 4 years and we released our terrain tech demo back in 2009 and the terrain rendering still looks much better than unreal and this is one of the first things we implemented.

Sorry to be a bummer and I am not trying to complain just simply stating something that would be quite simple to implement and very efficient.

Well, you have source access…

+1 would really like to see this 4033 building in WM is horrible

,
Im the type of technical artist who sit in between artists and programmers and fall more on the art side. This is the job of a kick *** coder not an artist and I wouldn’t even attempt it. It is however sound advice I give having worked developing engines such as the SOMA engine on many released games… so epic can either take the advice or not but I simply had to mention it.

If you turn on PN triangles tessellation in your material you get a similar type of smoothing on your landscape. Additionally, this would be evaluated on the GPU instead of the CPU, so I don’t really see the need for this kind of additional interpolation.

Yes… WM is horrible…

Arnage if you never had Bicubic I guess you perhaps don’t know what your missing out on and I don’t think bicubic needs to be CPU bound either. As far as I have seen the tessellation doesn’t do much to smooth out the terrain itself(unless I need some height map image for that as well)
Not sure what you mean WM is horrible ? I don’t really have any issues building a 4033 size maps and regularly build multiple 16k maps for larger scenes…just not with UE4

Well personally I don’t have a Xeon Workstation with 128 gigs of RAM .

The difference actually isn’t that great when applied to a heightfield. Here’s a paper that compares them, with figure 18 most closely resembling landscape usage. (the ACC patches use bicubic interpolation) Only at very extreme angles do they start to differ more significantly.

If you specifically want the best interpolation I think it would be more valuable to directly go for Catmull-Clark subdivisions. Shouldn’t be hard to do with OpenSubDiv either and would be great to have for far more then just landscapes.