I am currently creating a small plugin to create parametric Cornell Boxes (with configurable dimensions, options, lighting setup, wall materials, colors and so on), so that I can study the behavior of Lumen at small scale. I’ve learned quite a couple of subtle things with this.
Here is a little game for you:
There is 3 images of the same sample Cornell Box (a classic one, with just a Rect Light, diffuse white/red/green walls and 2 cubes inside). One image has been generated using Lumen, another one with the Path Tracer, and the last one with Static Lighting.
But you don’t know the order. So, can you guess who’s who ?
Don’t hesitate to download the 3 images to compare them side by side, as it may be a little bit harder than you think
Just for a bit of additional technical info, and to help guesswork:
Wall and cube materials have a metalness of 0.3, specularity of 0.5, and roughness of 0.6. Only the color changes for left and right walls, the rest is white (0.9, 0.9, 0.9) ;
Light is mostly white, with a really slight bend towards yellow (1, 0.95, 0.9) ;
Lumen is using Raytracing with Hit Lighting for reflections, and Raytraced shadows ;
Static Lightmaps resolution have been greatly increased and compression disabled to reduce artifacts in Static Lighting.
I’d be curious to hear what are your answers !
Here is a small poll to track results (images are numbered 1 to 3):
Already voted but out of curiosity, for static lighting; Did you build lighting with CPU lightmass or GPU lightmass? And are you using reflection captures or deprecated raytraced reflections? Is the light using static or stationary shadows?
It is CPU Lightmass, there is no reflection captures or deprecated raytraced reflections. But there is a high resolution lightmap set and in Post Process Volume, I increased of the number of indirect lighting bounces to 5 (vs 3) and set Indirect Lighting quality to 2 (vs 1).
Also, there is no Ambient Occlusion on Static Lighting (and I think I should have added it, since this makes things a bit more difficult
The light is set to Static mobility for Static Lighting, and is set to Movable for Lumen and Path Tracer
Darn, this is tricky. Going to share my thoughts via spoiler tag:
Spoilers?
The lightmass is pretty easy to suss out because there’s no specularity happening, but the difference between high-end lumen and the path-tracer really asks for examination. Lumen has a slight yellowing effect as many know, so I was trying to figure out what image was more yellow. Simultaneously however, it also tends to add in those somewhat unnatural contact shadows under certain circumstances, and the more 'blue (to my monitor and I) image had contact shadows that looked worse. The noise was also a hard one, because lumen can demonstrate noise in circumstances but also resolve quite cleanly.
By and large, I’m confident the first one was lightmass, but beyond that I could be proven wrong. I’m guessing that a metallic ball in the scene could be a dead giveaway.
I think if nothing else, this thread can be another proof of the Lumen potential. At least in the state PathTracer is right now - (and without spoiling too much) I think Lumen can look even better than PT, at least with certain tweaks and MRQ and because in certain conditions the color and light values can be better balanced.
Not to mention the advantage it gives you to work in real-time and to be able to rapidly make tons of adjustments and instantly observe the results.
I’m just waiting for a couple of more people to vote, in order to have an interesting-enough sample, before spoiling the real answer
Indeed, one of the 3 possibilities is easy to find. I discovered a bit later on that I could have make it more difficult though by tweaking a few options. And yes I am a bit positively surprised by the quality of Lumen vs Path Tracer… unless there is some mirrors reflections. The quality difference is then obvious.
Interestingly the one I was most confident about is the third one being the path tracer. Lumen has very noticeable over-occlusion in the corners of rooms where walls (and other meshes) meet, and produces a warmer color temperature than the path tracer.
Until the OP mentioned that they placed no reflection captures and were not using raytraced reflections, I wasn’t sure the first was lightmass. Especially since GPU lightmass can match the pathtracer quite closely, potentially closer than Lumen if used with raytraced reflections.
By the way, I’ve released the Blueprints I used to setup various Cornell Boxes with different customization attributes (box dimension and configuration, lighting system type, colors/materials, etc.)
