Lightroom 12.3 - Denoise is there

[mcbroomf]

See:https://community.adobe.com/t5/lightroom-classic-discussions/denoise-ai-in-12-3/m-p/13749953#M319651
 24.4MP (R6MII) raws on my MBP take 12 seconds to process. Saving the new, linear DNG takes about a second, that's not using GPU on this end.

Hi Andrew,
Any further info both 1) known, and 2) can be shared, on the issue with the neural engine not being used during denoise?  I've checked the Adobe thread you posted.  No update nor anywhere else so far as I can see.  Particularly interested in;

How much faster might Denoise be when using NE as expected
Are we waiting for an Apple fix or a temporary Adobe workaround? (I'm reminded of the access to a Lightroom catalog on external drive that accompanied Ventura)

Thanks

Mike

[bobtrlin]

Have just tried Denoise and am impressed. Julianne Kost’s tutorial is good. Updates to LR these days make it worth the subscription IMHO.

Jonathan

I agree.
In light of the offerings from Topaz and DxO, this is long overdue.  For better or worse, many of us are effectively tied to Adobe, so it's gratifying that we are not forced into cumbersome external editing and more cost.  We must remember that Denoise is still straight out of the box.  There has been not one tweak as yet.  I expect there to be many in the weeks to come.
I'm running LrC on a 12th Gen Intel i7-12700K with just the integrated GPU.  It runs LrC perfectly.  However, Denoise has thrown a spanner in the works.  I am addicted.  Denoise runs but the iGPU takes 4 minutes at 100% to process a 20MB file.  That's too long.  I'm resigned to having to instal some thumping big discrete GPU but I'll wait a bit to see what Denoise tweaks come along and for Adobe to update advice on optimum hardware requirement in view of Denoise.  In the meantime, I'll be taking a few more 4 minute breaks and put a bit of money aside.

[Chris Kern]

For better or worse, many of us are effectively tied to Adobe, so it's gratifying that we are not forced into cumbersome external editing . . .

The ideal situation, for this and the other Lightroom Enhance features, would be to apply the adjustments to the raw file dynamically ("parametrically," in Jeff Schewe parlance) rather than producing a linear DNG—albeit one that is scene-referred.

Apparently, Adobe is working on this:

We’re even looking into ways to speed up the workflow by not needing to make a new DNG file.

One of Lightroom's most valuable attributes—at least, from my perspective—is that it allows the photographer to maintain a raw workflow.  Demosaicing should be relegated whenever possible to producing an output file format for printing or static display.

[headmj]

I was thrilled when this update was issued. 

I  own most of the available denoise software.  DXO v2 and now 3, topaz and topaz AI and on1 denoise.  I'm a bit of a software junky.  Early on I stopped using topaz and ON1 mostly because of the DNG files they returned. 

I have run some tests on about 50 files that I previously processed in DXO.  IN DXO2 I found my self frequently turning off global lens sharpening because the output looked over sharpened (crunchy!  ;-) ).  particularly bird feathers.  In DXO3 I have found myself using "soft" selection in the lens softness selection.  This in an attempt to avoid over sharpening particularly in the XD method.

I have found that adobe denoise does as good a job removing noise.  It also looks as good after it has been properly sharpened.  The new XD selection in DXO3 runs slower than adobe.  I have had no problems with excessive run times with adobe.  They come in at about the projected times.  In the case of both pieces of the software the complexity of the image seems to matter.

I will be using adobe denoise from now on.  It returns simplicity to my workflow and really does the job.  I will keep DXO around in case I get images that adobe doesn't handle well but will probably not upgrade.  It may not work better but I will be able to experiment.  (software junky)  No more renewals for Topaz.  I keep ON1 around for its effects but probably will not use their denoise.

Thank you Adobe.

[StephaneB]

Hi Andrew,
Any further info both 1) known, and 2) can be shared, on the issue with the neural engine not being used during denoise?  I've checked the Adobe thread you posted.  No update nor anywhere else so far as I can see.  Particularly interested in;
Are we waiting for an Apple fix or a temporary Adobe workaround? (I'm reminded of the access to a Lightroom catalog on external drive that accompanied Ventura)

DXO is not using the NE either, they state it creates colour shifts and they are working with Apple on the issue. It could be the same issue Adobe is facing.

[wcarlew]

I'm running LrC on a 12th Gen Intel i7-12700K with just the integrated GPU.  It runs LrC perfectly.  However, Denoise has thrown a spanner in the works.  I am addicted.  Denoise runs but the iGPU takes 4 minutes at 100% to process a 20MB file.  That's too long.  I'm resigned to having to instal some thumping big discrete GPU but I'll wait a bit to see what Denoise tweaks come along and for Adobe to update advice on optimum hardware requirement in view of Denoise.  In the meantime, I'll be taking a few more 4 minute breaks and put a bit of money aside.

I am using an older i7-9700 (32gb ram and 1tb SSD) with a Quadro P1000 4gb, the denoise AI would take 4.5 min for a Sony A7R3 42mp file (estimate showed 8min). Also LR CC was very slow in general. A friend with a new Mac Mini M2 pro took 38sec for denoise AI with the identical 42mp RAW file.  I really didn't want to upgrade my whole computer so I took a chance and installed a GFX 3060 12GB OC video card.  LR CC is now very quick and denoise AI processes an A7R3 file in 20sec and 24mp RX1 files in 10sec.  I was very surprised the jump with just the video card, I would have been happy with sub 90sec. 

[mcbroomf]

I am using an older i7-9700 (32gb ram and 1tb SSD) with a Quadro P1000 4gb, the denoise AI would take 4.5 min for a Sony A7R3 42mp file (estimate showed 8min). Also LR CC was very slow in general. A friend with a new Mac Mini M2 pro took 38sec for denoise AI with the identical 42mp RAW file.  I really didn't want to upgrade my whole computer so I took a chance and installed a GFX 3060 12GB OC video card.  LR CC is now very quick and denoise AI processes an A7R3 file in 20sec and 24mp RX1 files in 10sec.  I was very surprised the jump with just the video card, I would have been happy with sub 90sec.

I've just ordered the same GPU to upgrade my system (an older desktop with an i7 7800 and a Quadro P2000 / 5GB).  I've seen several posts that suggest similar Denoise times for 3060, 3050 and 2060 GPUs while there's not a huge step up in performance for 40 series.  Given the amount of high ISO captures I make this should give my system enough legs to last through 1-2 more rounds of software development and maybe Apple resolve the NE issue to see if a newer Studio or updated PC might be my next upgrade.

[digitaldog]

In terms of Denose (or any NR) and high ISO, can someone explain why high ISO instead of underexposure is always in the mix when people speak of noise?
An example made this morning:



Details:
Canon R6 Mark II.
ISO 200 (yes, I can set it lower with a custom setting, no need).
Meter: Minolta Flash Meter III incident reading.

IncidentMeterNormal.dng:
Normal exposure is recommended exposure of the Minolta meter: 1/15th@F8
In-camera meter tells me that would be 1 stop overexposed. It recommends 1/30th@F8.


-3stops.dng
125th@F8 or three stops below incident meter recommendations: UNDER-EXPOSED.
All settings in Lightroom Classic’s Detail OFF.

Does all this talk of 'high ISO' take the new (or older) NR off the radar for some users?

URL of screen captures if that makes it easier to view: http://digitaldog.net/files/ISO200Normal_minus3stops.png

[digitaldog]

One more test.

1. ISO 800 but using above-recommended exposure 15th@F8.

2. ISO 50 but 125th@F8. 

Detail settings in Lightroom Classic OFF.

Low ISO produces less noise: No. 

High ISO produces more noise: No. 

[fdisilvestro]

In terms of Denose (or any NR) and high ISO, can someone explain why high ISO instead of underexposure is always in the mix when people speak of noise**?

Those are the people who believe that ISO is part of the exposure.

[digitaldog]

Those are the people who believe that ISO is part of the exposure.

+1. It isn't.

[jrsforums]

One more test.

1. ISO 800 but using above-recommended exposure 15th@F8.

2. ISO 50 but 125th@F8.

Detail settings in Lightroom Classic OFF.

Low ISO produces less noise: No.

High ISO produces more noise: No.

In this case, to prove your point,  wouldn’t you want the exposure in both cases to be the same?  That is case #2, ISO 50 with 1/15@f8.  Then, assuming a reasonably linear sensor, both images would show same noise….that is, no change due to ISO change

[digitaldog]

In this case, to prove your point,  wouldn’t you want the exposure in both cases to be the same?  That is case #2, ISO 50 with 1/15@f8.  Then, assuming a reasonably linear sensor, both images would show same noise….that is, no change due to ISO change

This is a test to show ISO alone isn't the cause of more noise. It is a test to look at high ISO with automatic exposure (see below, sorry, another copy and paste):

ISO does not affect exposure. ISO affects an exposure recommendation. You (can) control exposure if you wish.

Setting ISO speed does not change the sensor's sensitivity to incoming light like volume control does not change the sensitivity of a radio. In both cases the setting (ISO or volume) controls only the signal processing, while the input stage (sensor, antenna) provides the same input signal and in this case, exposure to the sensor.
That's why when ISO setting is cranked up, automatic exposure results in more noise - automatic exposure, in this case, decreases the exposure (that is, the combination of aperture and shutter speed is set to allow less light captured by the sensor): less exposure, less light, more noise.

[digitaldog]

In this case, to prove your point,  wouldn’t you want the exposure in both cases to be the same?  That is case #2, ISO 50 with 1/15@f8.  Then, assuming a reasonably linear sensor, both images would show same noise….that is, no change due to ISO change

Actually, I did this years ago and it IS worth considering:

The higher ISO with the same exposure has less noise in this camera.
I should also test this with my newer Canon.

[fdisilvestro]

I suggest that everyone do their own tests, but what I have found is that given a specific aperture / shutter speed combination, you will get the best results by increasing ISO as much as possible without clipping the highlights.

[jrsforums]

I suggest that everyone do their own tests, but what I have found is that given a specific aperture / shutter speed combination, you will get the best results by increasing ISO as much as possible without clipping the highlights.

From what I have read, that is generally the case, but, for most current sensors, negligible unless you are utilizing dual ISO sensor.

[wcarlew]

I've just ordered the same GPU to upgrade my system (an older desktop with an i7 7800 and a Quadro P2000 / 5GB).  I've seen several posts that suggest similar Denoise times for 3060, 3050 and 2060 GPUs while there's not a huge step up in performance for 40 series.  Given the amount of high ISO captures I make this should give my system enough legs to last through 1-2 more rounds of software development and maybe Apple resolve the NE issue to see if a newer Studio or updated PC might be my next upgrade.

Hope you see a significant improvement as well.

[mcbroomf]

Hope you see a significant improvement as well.

30 seconds for a 50mp Sony A1 raw, vs 3-4 minutes before.  Pretty happy with that until my whole system needs to be replaced.

[NikoJorj]

The higher ISO with the same exposure has less noise in this camera.

Welcome to the Canon club! This high read noise is a perk of Canon sensors (among others).
With a perfect (ISOless) sensor, noise should be the same with the same exposure (counted in photons).

[digitaldog]

Welcome to the Canon club! This high read noise is a perk of Canon sensors (among others).
With a perfect (ISOless) sensor, noise should be the same with the same exposure (counted in photons).

Yes, indeed (although I haven't tried the same test with my new R6 Mark II. It Sounds like you're saying don't waste the time). At least with my old Canon, it isn't ISO invariant.

I've been working on a (short?) paragraph or two on this; comments are welcome to get it smoothed out:

The term "ISO invariance" refers to a camera's ability to maintain consistent image quality while capturing images at different ISO settings. Suppose a camera is (perfectly) ISO invariant at all ISO settings. In that case, it will produce the same level of image quality whether the ISO setting value that was set at capture or image brightness is adjusted in a raw converter. This ISO setting “doesn’t matter” in that scenario.

A non-ISO invariant camera may introduce more noise or less noise depending on the exposure due to this ISO setting; there is a variant to the outcome of setting the ISO.
In both kinds of systems, ISO will affect the recommendation of exposure (Aperture + Shutter speed), which affects the result of noise. In a non-ISO invariant camera, a higher ISO at optimal exposure can produce less noise than a lower ISO setting using the same exposure! This should not be the case with an ISO-invariant camera system. Hence the name and behavior. This behavior destroys the myth that “higher ISO produces more noise.” Underexposure is the result of (increased) noise, not ISO.
Altering ISO and, thus, the exposure recommendation can come in handy in low-light situations where you may need to increase the ISO setting and this exposure recommendation to avoid camera shake or stop motion with a higher shutter speed. Or if you wish to alter DOF, that wouldn’t be recommended* at a lower ISO setting. This results in less exposure and more noise. Again, it is worth stressing: underexposure is the cause of (increased) noise, not ISO.

* Lastly, the ISO provides a ‘recommended’ exposure (Aperture + Shutter speed), but the savvy photographer does not need to accept this! Even without the inclusion of ISO, a savvy photographer will often ignore a reflective meter recommendation, for example, a black cat on coal or a white dog on snow. Such a recommendation would produce an overexposed or underexposed capture, despite the ISO setting in those examples.