On Filtering the Noise from the Random Parameters in Monte Carlo Rendering

[Patan77xD]

really cool technology,

but you are doing something seriously wrong because that would't take a year to render...

[yousdarabi]

We rendered these videos with 8 core machine and that's why we called it high -end and agree that in industry they are using render farms and therefore it takes much less time but our approach can be accelerated with the same rate because it is completely parallelizable.

[mjohansify]

These guys used PBRT (a slow rendering system) to render their images, and were tracing 8000 paths/pixel (to get the ground truth) which is why it took 24 hrs to render. The point of this work is that they can accelerate the rendering process more than 100x and still get good results. Their results are with only 8 paths/pixel! So even if you use a GPU path tracer or a render farm, you still benefit from this algorithm by tracing less rays to get a good result. That is the point of the paper.

[Cyostasys]

Well, I wont call the persons made a realtime pathtracer amateurs ;-) but yes, I also cant believe that an image at this lightsitution would need 24 hours to render noisefree. But I have to say, that the result from the filter is quite impressive

[otonanoC]

***Each reference frame took a little over 24 hours to render on a high-end machine.***

I don't believe you. I thought you academics had access to render farms. Surely you've seen what a GPGPU can do for path-tracing. Even amateurs have made path-tracing renderers that run in real time.