Tuesday
Feb282012
Denoiser II
Tuesday, February 28, 2012 at 11:18AM 
From the Red Giant blog:
- We’ve developed Magic Bullet Denoiser II from scratch (completely new code) to bring you the quality of Denoiser 1, but with more stability. Red Giant owns this code 100%—which means the product is not going anywhere.
- We have spent lots of time developing the default settings for Denoiser II to give you the best possible look as soon as you apply. There’s is always room to tweak setting, but often you won’t need to.
- Currently Denoiser II is only available for After Effects, but it will support multiple host apps in the future (read this FAQ for more info).
- Denoiser II is a Free Update for all Denoiser 1 and Magic Bullet Suite users. I repeat: If you already owned a license of Denoiser or bought the Magic Bullet Suite after Denoiser was discontinued, you get Denoiser II for FREE.
I use this on everything I shoot.
Reader Comments (23)
Excellent! Downloading this now. Kudos for redeveloping it from scratch and for making it available to existing users for free. I hope a Premiere Pro version will follow very soon!
The last line of this post carries such weight. :) Actually I would like some clarification on that, is really everything you shoot, even at low ISO levels?
I will have to give this a try.
Dave
Dave, it's a great question. Denoiser is not just for "rescuing" high ISO footage. It also cleans up compression artifacts and helps bump up the bit depth of 8-bit footage. So I use it on everything.
great news!
and thanks for caring so much about us, old customers
Timely post.
Perhaps of interest is I recently (Monday) did some testing of denoise software as I have someone who has shot a a 12-day, 12-night dark sky timelapse for my IMAX film with a 5 camera rig (5D Mark II). There are 120,000 raw files to denoise.
I tested Red Giant, RevisionFX, Neat Video and Noise Ninja (which is Photoshop only). Both Red Giant and Neat had unacceptable banding on the twilight shadow gradients. RevisionFX and Noise Ninja did not. I did try various settings. Is this user error on my part? And/or is there some code in the new release to deal with this. Note all processing was done in 16-bit on identical files.
Stephan, I would love to see some screen grab comparisons if possible :)
Did you use a sample area or just put in default?
How does this compare to things like Dark Energy?
http://www.cinnafilm.com/
I know Dark Energy was used on Act of Valor to great success. But it's pretty expensive...
Stephen, sometimes when super-clean 16-bit gradients suffer from display banding. The banding isn't actually in the footage, it's happening at the 8-bit pipeline to your display. The way to check this is to add a tiny bit of noise. If it's display banding, it will change and gradually disappear as you add noise.
I always add a little noise back in to denoised footage. Even if there's no banding on your display, there very well might be when you compress for the web, or broadcast, or whatever.
Mike, the Dark Energy system has this re-graining component built-in and it's a big part of how they market it— mimicking real-world film stocks and stuff. But it does start with removing noise.
Could the banding come from your 'Previews' settings not being set to 'More Accurate'
Stu:
it's definitely not display banding and adding noise is a technique I use - no effect here - and you see in 10 and 12 bit output on other display. Plus the photographer on his system (two Macs) noticed the same issue. I've talked to several astrophotographers and they have run into this issue.
It may be something we're doing though - so hopefully you can shed light on it.
Marc,
I will get a couple of samples up - and I always use "more accurate" in AE for just that reason as most of my work is destined for larger screen output.
Would love to see those examples. We're always looking for ways to improve the technology!
Okay, I don't trust downsampling, so I've uploaded a simple AE test project (you'll need revisionfx/MB plugs installed, CS5.5. Also are 16-bit PSD output that on everything I have here shows the banding. Plus the source RAW file is there.
It's here: http://www.outsideinthemovie.com/collab/noise/
Let me know what you see and if we are doing something wrong.
FYI - this one of the 120,000 frames taken by this rig here:
http://vimeo.com/31068829
If you would like to a sample timelapse, email me and I can see offlist as photographer has it protected.
Thanks Stephen! I started by loading the magic bullet.psd into a 16bpc projects in After Effects and I can definitely see the banding. This is a challenging example! But it is definitely happening at the display, not in the file itself. The way I can tell is that the mach bands move when I adjust the exposure of the Comp window.
The banding completely disappears when I add 1% noise using the After Effects Noise effect. In fact, this image is about as good a demo of that trick as I've ever seen.
Another good way to experiment with this is to use Levels to crush the contrast of the image. If the banding is burned in, it will remain static and become pronounced. If not, it will move around and be reduced.
So... why does the Denoiser example band when the RE: Vision Effects one does not? It looks to me like the RE: Vision Effects plug-in is actually adding some noise back in to fight banding. The Levels crush trick shows this very well. Smart! Maybe we should consider doing that in Denoiser—although we'd be sure to make it optional.
The Noise Ninja sample seems to be in a different color space, so it's hard to compere precisely, but it looks like the same thing—fine, pixel-level noise eliminates the banding.
Let me know if your findings are different. Like I say, we're always happy to find ways to improve. Thanks so much for taking the time to send it!
I was able to correct the color space disparity (something with Adobe RGB and ProPhoto) and make a direct comparison among all three. You can view the results, crushed with Levels, here. Denoiser II on top, then Re: Vision, then Noise Ninja.
If you add 0.5% noise to the Denoiser II image before the crush, it looks exactly like the other two. You can try this yourself using the After Effects (CS5.5) project here.
One more thought on this: One reason we don't add noise to Denoiser's output (aside from the obvious) is that After Effects itself dithers 16bpc images when rendering to 8bpc formats. You can see this happen without having to render by adding an 8-bit effect to a 16-bit image in a 16-bit comp—like the Timecode effect, for example.
Remove the Noise effect and add Timecode to the magic bullet.psd layer in the sample file and the banding disappears as well.
Stu - thanks so much for looking at this and the examples. I see some of what you are saying but other stuff. When I drag the images in Photoshop, the banding moves with the motion. However, as you point out, adding noise does help but on my calibrated wide gamut I was not adding enough - I need around 2 percent before it goes away and I only added upto 1% before concluding it was not display banding.
But, if in Photoshop, i crush the levels, the banding does vanish at 100%. Curiously, with the levels crushed, when scaled down, there is strong display banding in the Magic Bullet, slight in Revision FX and none in Noise Ninja.
That - and the amount of noise needed to minimize tells me it may not be all or nothing.
Display banding may be the larger part, but the the noise reduction method is also a factor in creating differences triggering display banding. Is it possible this related to how smoothly the noise reduction is applied. Using the info, I can't tell scientifically, but it does seem the steps/curves as the least smooth across the processed image in MB, medium smooth in RevisionFX and smoothest in Noise Ninja. Or am I imagining things?
Anyway, you are much more knowledgeable about this and thanks for clarifying. I was not adding enough noise to banish banding on my setup. I do have mixed feeling about adding that much noise in prior to noise reduction - are you sure that's what Noise Ninja & RevisionFX are doing? Is there someway to test (to force them remove noise on an noiseless image to see if any is added in)?.
Couple of other notes, when I rendered in AE, MB failed multiprocessing in this project by RevisionFX did not - maybe something to report to engineering.
Again, your time and expertise is great. Wish I could afford to pay you to be a consultant on this project as unlike even film work, when projecting on 8 story high screens I've been stunned how much "every pixel matters" and lots of tricks that work fine, fail at those screen sizes.
Not prior to, after.
Not at all. You're just seeing the results of adding noise. Denoiser adds the least (none), Re: Vision Effects adds some, and Noise Ninja adds the most—at least, to look at these specific samples anyway.
As Tim Sassoon once said, "Noise is image vitamins." Add it until you're happy, testing your output through the pipeline its ultimately going to take.
>Not prior to, after.'
Sorry brain fart. I'm EST so it was late here.
>As Tim Sassoon once said, "Noise is image vitamins."
Tim helped me get started in getting connected to the IMAX/Large Format folks including Rick Gordon etc. at RPG and Andrew Oron at Fotokem. But one issue I've discovered is their recipes including Tim's are based on experience going out to film and/or having film sources. My composites are often clean and with sudden recent changes in IMAX digital projection, I'm optimizing for digital now.
This is why I'm currently trying to figure out how to panoramic stitch this sequence in AE instead of a PTGui or similar. Noise reduction etc. can stay live, one optimized for film output, one for digital instead of baking it in (as recommended by still stitchers) for 120,000 raw files (almost 4 TB's worth).
Stu, on average how much noise are you adding back in? Never really thought about doing it much. I;ve been setting it around 2% and that seems good, but not really noticeable in most shots.
That's exactly what you want—just enough to not be noticeable.
Hi Stu,
You say you use it on everything you shoot. How?
I am very new to AE so perhaps the problem is my workflow but... I don,t want to use it on everything, just a simple clip, 1:30s long. So I selected it in premiere, chosed to replace it with an AE comp. In ae, I dropped Denoiser II on my clip, result seem great, So I added my comp to the render queue. Just to render 1:30, It took 68 minutes...
So how can you use it on everything? In this particular case, I can wait this time, the plugin saved this particuliar clip. But on everything? How Stu?
By the way it was an AVCHD 30p clip, 1920X1080, The only effect added is Denoiser.
I have a somewhat good computer, a PC with a I5 2500K CPU, and 16 GB of ram.
That render time is crazy too long. Something's up. Please contact Red Giant support.
Hey Stu,
I've got a question (or two) about my workflow as it pertains to denoising and color correction. I like the DV Rebel onlining process, but, it's not practical for me since my work needs to be very fast paced. I shoot weddings (all DSLR), so, I'm in several different rooms with different lights and nearly 100% practicals.
My workflow is simple: I do my editing in Premiere Pro (CS5.5) and I prefer to do my Color Correction in Premiere as well because of the built in scopes, and according to Adobe, Premiere works in 16 bit now, it's a faster workflow for me and I have some after market transitions that I use frequently..
I correct and grade with Colorista II and use Looks to punch it up on occasion.
So, here is my dilemma, I prefer to denoise all of my clips after editing and before color correcting and grading.
So, question number 1 is: Is my understanding correct that Premiere Pro now processes in 16 bit and is therefore better for coloring than before?
2. Is there a good way to round-trip an edit to AE for denoising then back to Premiere for coloring?