New "producing" filters: noise generators

Here is a package containing two new "producing" filters, UniformNoise and GaussianNoise. As you can probably guess, these respectively generate uniformly distributed noise and normally distributed (aka Gaussian) noise. Their usage is vaguely similar to the built-in BlankClip filter. Please check out the included readme for more information. (See for the reason I wrote them, and for some discussion of the benefits of adding noise to a clip.) This is first-release alpha code, and the usual disclaimers apply. Feedback of any nature would be very much appreciated. Edit: Removed link to old version.

Here's (and ), which fixes a pure virtual call runtime error when temporal=false. I also put up a with the filters I've written to date. It's quick and nasty but it works ;).

Hey, is that a tumbleweed? *grin* Incidentally, I just finished a test encode of American Pie with this script: Code:

GaussianNoise is sloooow. Painfully slow. 32 bit precision is ludicrously overkill when you're just outputting 8 bit noise. Back away from the floating point ops. MINMAX(n,16,235) == bad juju. Don't assume people are using the CCIR601 color scale.

Your filters look to give many benefits,especially for those outputting to TV. I use cce to encode svcd for tvout and never knew what the 16-235 setting did until reading your LegalClip readme..thankyou.I will use it for my divx rips that i output to TV from know on,looks much nicer.

GaussianNoise is sloooow. Painfully slow. Of course -- it's unoptimized alpha code ;). Apart from that, transforming uniform noise to Gaussian noise is never going to be particularly fast by it's nature. I use about the fastest algorithm out there, but as you observe it's not speedy. That said version 0.3 has a new parameter, loop. This specifies the number of frames to generate which the filter will cache and loop over. For example, if you say loop=6 it'll generated and cache six frames of noise the first time it's called, and will simply loop over those for the remainder of the clip with no more noise generation to be done. In certain applications (e.g. adding noise to a clip) this seems to work just as well as "proper" noise but has the advantage of running at real time. Back away from the floating point ops. Unfortunately I'm not sufficiently mathematically inclined to convert those floating point ops into integer-only versions. If anyone can point me to a good resource to read, I'd be grateful. In fact I made the decision to use floats instead of doubles for performance reasons, but I'm now considering switching back to double precision. It's amazing what damage an "innocuous" change can do to a PRNG, and I'm nowhere near close to understanding the ramifications of switching from double->float. Now, with the loop parameter, speed is much less of an issue for what I perceive as the most common application for this filter. MINMAX(n,16,235) == bad juju. Don't assume people are using the CCIR601 color scale. I actually started a trying to address just this question. At the moment there seems to be a great deal of confusion (at least in my head ;)) over whether or not, as far as Avisynth is concerned, YUY2 should always be within CCIR-601 bounds. Most existing filters do clamp.

Your filters look to give many benefits,especially for those outputting to TV. I have to say the resulting encode of American Pie far exceeded my expectations in terms of quality, especially since the avs looked so noisy in Media Player. But it seems that MPEG-1 takes that noise and smooths it significantly, and in this instance it was nearly invisible in the final encode. So I'd recommend encoding a short clip with various "strengths" of noise and see what the encoder produces from it, rather than simply trying to get the "right" amount of noise by looking at the avs in Media Player or whatever. I use cce to encode svcd for tvout and never knew what the 16-235 setting did until reading your LegalClip readme..thankyou.I will use it for my divx rips that i output to TV from know on,looks much nicer. Yep, it can make a big difference, since by scaling to 16-235 you're preserving all that detail that is otherwise clipped by your DVD player or TV out card...

Thank you so much - I've been waiting so long to get one from Donald Graft; I'm sure he had much more interesting things to do, though...

Thank you so much - I've been waiting so long to get one from Donald Graft Wow, someone else who finds it useful! :D Out of interest, what did you need a noise generator for?

What I would need a noise generator for? Well, for testing purposes, of course (codec quality, for example); for special effects, maybe; and even for enhancing the quality (you know, forcing a few high frequencies may reduce blocking artifacts).