ompf2.com

This is quite spectacular:

http://agl.unm.edu/rpf/

Moreover, it's very likely that it can be brought near to real time on a high end GPU or two, since it looks like a very GPU friendly algorithm, I'd expect a speed increase of 100x or more, since it's heavy on local number crunching and uses some transcendentals (which are very, very slow on CPU, but effectively full float rate on GPU, so long as no more than 10% ish of all instructions or less are transcendental). Moreover, there are some very effective approximate algorithms for things like the bilateral filter...

Anyone have a nice easy to work with CUDA path tracer that I can experiment with? I'd like to try coding a fast implementation of this algorithm, but it does require some minor modifications to the path tracer itself, namely outputting various intermediate values to other frame buffers, much like a typical deferred renderer.

You can use Brigade2 for that. I'll gladly provide source code for research purposes. Drop me a line at bikker.j@gmail.com .

You can also use the Dade's SLG code http://src.luxrender.net/luxrays/

The simplest is to use a framebuffer with all the data, load the data from a file and then apply your algorithm. I don't remember where but I have see the same kind of algorithm coded in CUDA and C++, but without renderer.
This way you can easily plug it into any renderer

Be carreful, there is a commercial license for this technique !

Here is the article I was talking about, it is a related subject :

Temporal Light Field Reconstruction for Rendering Distribution Effects : http://groups.csail.mit.edu/graphics/tlfr/

And the code is provided

Keep us informed of your progress

It seems spectacular

spectral wrote:You can also use the Dade's SLG code http://src.luxrender.net/luxrays/

I have still to read the paper but if it as promising as it looks, I'm going to try it for sure.