ompf2.com

I'm not sure thresholding would help. MLT already favors bright paths, and allowing mutations to dim paths would reduce the probability of getting stuck in a small region of path space. I think discontinuities would be noticeable as well.

Typically normal maps are derived from bump maps. It's essentially a precomputation step that saves you from having to calculate surface derivatives at render time, which can be overly expensive when you only have a dozen or so milliseconds to display a frame. On the other hand, normal maps aren't v...

Used naively, probably not, but there are advantages to amortizing the cost of light paths over many eye paths. IGI maps well to GPUs--you path-trace light paths, but can use rasterization and shadow maps instead of eye paths and shadow rays. If you have a lot of VPLs, you can prune the set judiciou...

The issue I find is that my images appear to get brighter when I use importance sampling due to the PDF always being less than one (well less than Pi) and therefore increasing the radiance when used for division. Surely when sampling the PDF the value should occasionally equal more than one so that...

The image is quite dark, at least on the devices I have handy, so I can't comment on the artifacts themselves. How good is your random number generator?

sorry for jumping over the thread, but I wonder how ERPT works? my previous knowledge so far, how PT works, BDPT, VCM, PPM, PM + final gather, MLT (BDPT), MLT + MIS (MMLT) how ERPT is different? from the name it seems to apply only to PT, but can it apply to BDPT or VCM like algorithms? does it con...

Hm. I wonder if there's some nutty MIS way to automatically combine "regular" path tracing and MLT... Well, in a way that is what Keleman is doing by mixes large step mutations with small step mutations using MIS. Basically it mixes MLT and PT using MIS. MLT is best at sampling high contr...

On another note, whether MLT really produces higher quality samples is debatable. Sure, the sampling density is proportional to the target function, perfect importance sampling and all that, but it comes at a price. When MLT gets stuck in a peak of the target function for too long the correlation a...

Perhaps G3D (http://g3d.sourceforge.net/)?

Very interesting. You need external buffers to store histogram information at each pixel, but still good results and I think this is fast enough to do in real-time :) Unless I'm misreading it, the paper cites filtering times ranging from 5 to 115 seconds, which seems rather large for real-time appl...