I am subscribed to the Disney research channel, and I see really cool stuff there all the time. its usually a bummer though when I find no public repo or code is available for their stuff. Every time feels like blue balled....
Disney doesn't share. It's an ad, nothing more, sadly. It's worthwhile reading on Disney and IP. It's like Microsoft on software but way broader. Arguably impressive research results but with sadly only motivate by capturing value.
Neural proxies are a constant-time approximator for anything.
There's still difficulty in finding exactly where the proxy should go, i.e. what step can be approximated without losing fidelity in the output. Apparently, you also need to select auxiliary features to guide training. But if you figure those out, you can replace hours of computation with milliseconds, accurate to the limits of human perception.
The new model can be queried in new ways. The encoding from the RenderMan scene into the neural proxy allows for differentiable relighting. Once you have differentiability, you can find out what inputs create a desired output with gradient descent, instead of trying every possible input.
Neural proxies are also much, much faster for only a small drop in fidelity. 1 hour to 60hz is a 216,000x speedup. That's not possible without neural proxies. You could try lowering precision, resolution, bounces, scene complexity. Accuracy would be out the window before you get close to that performance.
It requires a fairly expensive precomputation pass and can only work for static scenes.
Meanwhile interactive path tracing is fast enough that the scenes they showed would only be minorly slower to be truly interactive with dynamic scenes.
I wish they’d showed this with scenes that don’t fit in GPU memory so it could show the benefits for CPU only renderers, otherwise GPU based renderers would be fairly fast with these scenes.
The only big thing for me was the multi view lighting. The painted light to light parameters is a neat trick but been done quite a few times in the past with traditional techniques too.
I am subscribed to the Disney research channel, and I see really cool stuff there all the time. its usually a bummer though when I find no public repo or code is available for their stuff. Every time feels like blue balled....
Disney doesn't share. It's an ad, nothing more, sadly. It's worthwhile reading on Disney and IP. It's like Microsoft on software but way broader. Arguably impressive research results but with sadly only motivate by capturing value.
It's an ad for people, for staff capable of that type of work. Very few can do that type of work these days.
Yep, staff is technically brilliant, company itself is still terrible though IMHO.
Feels good to read this :)
These days with LLMs, can't you get just ask the AI to do a reproduction if you really like the research?
Currently no, not at all.
If you are skilled in the field already, yes.
Though it's not just "ask" in the one-shot prompt sense, but more in the "over the course of a few days" sense.
Neural proxies are a constant-time approximator for anything.
There's still difficulty in finding exactly where the proxy should go, i.e. what step can be approximated without losing fidelity in the output. Apparently, you also need to select auxiliary features to guide training. But if you figure those out, you can replace hours of computation with milliseconds, accurate to the limits of human perception.
Aren't they a faster, lower fidelity model, like all models?
It is a faster, lower fidelity model.
but it's not JUST a faster, lower fidelity model.
The new model can be queried in new ways. The encoding from the RenderMan scene into the neural proxy allows for differentiable relighting. Once you have differentiability, you can find out what inputs create a desired output with gradient descent, instead of trying every possible input.
Neural proxies are also much, much faster for only a small drop in fidelity. 1 hour to 60hz is a 216,000x speedup. That's not possible without neural proxies. You could try lowering precision, resolution, bounces, scene complexity. Accuracy would be out the window before you get close to that performance.
https://www.eigensteve.com/
https://www.youtube.com/@Eigensteve/search?query=pinn
I found this paper quite disappointing.
It requires a fairly expensive precomputation pass and can only work for static scenes.
Meanwhile interactive path tracing is fast enough that the scenes they showed would only be minorly slower to be truly interactive with dynamic scenes.
I wish they’d showed this with scenes that don’t fit in GPU memory so it could show the benefits for CPU only renderers, otherwise GPU based renderers would be fairly fast with these scenes.
The only big thing for me was the multi view lighting. The painted light to light parameters is a neat trick but been done quite a few times in the past with traditional techniques too.