Convergence in high contrast scenes

[galinette]

lyc : Off course, I wanted to say "is more effective, as the scene uses highly indirect lighting". Here light must enter the light pit, then the windows. In that case QMC casts a high ratio of rays ending nowhere and not contributing to the image. This was not a general statement!

Initially, the spirit of MLT was even to avoid the use of portals as an ultimate goal : in the original paper from SIGGRAPH 1997, they pretend that MLT is able to find and lock difficult light paths without external help (such as portals)

Etienne

[PureSpider]

I know this is a bit off-topic but I've always wondered what rendering mode would be best for the images I do (those ones).
I'm currently using MLT BiDir but always wondered if PT BiDir or even just PT would be faster and still keep the caustics.
In general: Are those images simple or tricky scenes?

€dit: The images are lit by HDR only.

[galinette]

PureSpider:

Best is to try... There are no really sharp caustics in these, so MLT is maybe not mandatory. However, I do not think that MLT slows the render that much. To me, the drawback of MLT is for quick preview, as the noise is made of big strokes, you do not have a good idea of the final image. PT gives smooth noise which is better to figure out what the final image will look like. But I'm not sure that in simple scenes MLT is slower.

And also, you seem to use envmaps as the only lighting (am I right?). In that case BiDir is often causing more overhead and not helping much. In your case, if you render objects under an envmap, I would recommend turning of BiDir. To me, it only helps when you use concentrated lights, and slows down a lot otherwise.


So my (personnal, maybe wrong) recommendation for your images would be to use MLT without BiDir. If you want a keep preview with noise, use PT.

Etienne