Optimizing LPV for fully static geometry

Development Rendering
1

Hi all,
First time here, and very new to Unreal, so my situation is a mix of unfamiliarity with Unreal and a fairly specific rendering need. Also not sure if this belongs in the forums or the AnswerHub…

I’m trying to create a walkthrough of an architectural project with a focus on an accurate representation of atmospheric daylighting. Indirect lighting is key, of course, and I’m more interested in approaching physically accurate rendering than in artistic control of the lighting. There might be some static artificial lighting in the building, but mostly I’m concerned with sunlight, which needs to be able to change direction. The geometry can almost certainly be completely static. I’d also like to be able to use the Oculus Rift for experiencing the walkthrough, which has limited somewhat the tools I can use, and is part of the reason for trying Unreal. It’s only going to run on one computer so I’m not concerned about scalability. That said, I don’t have the top of the line, so performance is unfortunately still part of the discussion.

I’ve had two thoughts about how to make this happen, but of course the main goal is to get the above working and the questions below are only the ones I think I need to ask! Perhaps I’m headed in the wrong direction.

The first option, I think, is to use LPV’s for fully dynamic lighting. I’ve read the tutorials and gotten this working, more or less, just a few questions. Are LPV’s using spherical harmonics? After reading what I could find, my basic understanding was that SH’s were something you could pre-compute, and then quickly evaluate at run-time. However, in Unreal the lighting seems to be calculated as I move around - for example corners are black when I turn to look at them and then gradually brighten after a few seconds. Is this because the geometry is assumed to be potentially dynamic? Is there a way to get better/more accurate lighting by somehow telling the engine that it doesn’t need to worry about the geometry changing (Just shooting in the dark here)? Finally, are the default LPV settings the most “accurate”, or do the most physically accurate values for those controls change depending on the scene?

Another option I thought of was to bake the lighting using 3ds Max. Since the lighting changes, but the path of the sun is predictable, I could create a video from a sequence of bakes throughout the day. I could then use this plugin to display that video as a texture on the building. Is that feasible? This would be equivalent to having a Lightmass bake as far as accuracy goes, I think? Would I then use environment probes to create reflections?

Not sure if it’s ok to talk about other engines, but a related question is whether Unreal is the best tool for this. It feels like there is a lot of Unreal that I’m not making use of (e.g. ALL interactivity, except moving around) that might make Unreal’s focus different from what I need. Maybe something specifically set up for static geometry, like Renderlights would be better optimized? I haven’t gotten a chance to test out that engine yet though.

2

Oculus support means you have a combination of heavier rendering with a need for very consistent high framerates. If you want to achieve that in combination with accurate lighting you should forget about LPV. (or any kind of realtime GI)

The reality is that every realtime GI solution in every engine you’ll find will have the following characteristics:

  • Significantly heavier than precomputed lighting
  • Far less accurate then precomputed lighting

They might be worth it in some cases, but for your use case going fully dynamic does not seem to be worth it.

Using animated textures could work to a certain extend, but will make your process a lot more complex. In addition all you reflection will be broken as reflection probes are static and won’t be updated, while screen space reflections are not supported in combination with the rift.

Here’s how I would do this:
Statically light your environment with Lightmass. This will give you the best possible visual quality/performance. Now make a number of copies of your level and adjust the lighting to various different situations you’d like to showcase. Make all of these levels sub-levels of a single persistent level and load/unload these sub-levels whenever you’d like to switch lighting. As the geometry is all in the same location you don’t even have to translate the player’s location between the sub-levels, so this should be very easy to set up.

This will not allow super fine control over the time of day, but I assume your clients will be more interested in seeing the difference in lighting between midday and dusk than they are interested in the difference in lighting between 6:00 and 6:05.

Finally this approach makes it trivial to add sub levels with other potentially interesting variation such as what the lighting would be like in different weather conditions. Toggling indoor lighting also no longer requires more expensive realtime lights and can be baked into the night sub-levels but turned off in the daylight levels.

If you don’t like the instant transition (pop) this would create, you could mask it with some kind of short fade in between the different sub-levels.

3

Thanks for the ideas! Unfortunately I don’t think the multiple sub-levels is a route I can take. This is actually for a thesis project where I am exploring the way that light moves through a space. (Some related examples are Newgrange and James Turell’s Roden Crater.) Smooth transitions are very important - more important, actually, than reflections, though ideally of course I would have both. I think I’ll look into animated textures, but if anyone has other ideas, I’d love to hear them.