R5 and R6 noise reduction?

At what ISO would consider using noise reduction like Deep Prime for either of these cameras? Thanks!

As Deep Prime only supports RAW files, any ISO. ;-)

I'm not experienced, but I doubt that it's very useful much below 800.

Phranc said:

At what ISO would consider using noise reduction like Deep Prime for either of these cameras? Thanks!

Click to expand...

I 'd not consider making a plastic fantastic at all, but applying NR depends FIRST OF ALL not on ISO but on exposure... properly exposed "high" nominal ISO requires less NR than badly underexposed "low" nominal ISO (or vice versa) - more so when camera's sensor actually delivers less read-related noise at higher ISOs (like dual-gain sensor or older Canon sensors with ADC off chip... excluding cases with intermediate ISOs)

Phranc said:

At what ISO would consider using noise reduction like Deep Prime for either of these cameras? Thanks!

Click to expand...

Deep Prime is probably not the right kind of noise reduction algorithm. It's not their fault, it's Canon's fault. Take a look at the RAW noise in Canon cameras, it's not real. Either by virtue of some flawed de-bayering algorithm, or by intentional manipulation, Canon destroys the pure sensor readout and noise patterns. Of course, all cameras do this now which is why customers have been locked into these upgrade cycles. Not that it's all bad of course, getting better more capable gear is great, in theory. The assumption we all make is it really is better and more capable. But since they manipulate the sensor noise after the fact, they could be leading us down any one of their chosen rabbit holes.

Assuming noise is even a real phenomenon, it may not be, it would be very easy to remove using the same tech they're using for the AF. Machine learning. The key thing is that noise is not random, it's generally related to the input. Because of that, noise follows a pattern related to the image, machine learning could easily remove the noise by being trained on 10's of thousands of noisy images. The thought is, the algorithm could then see how the noise tends to follow the shape of the image and simply correct the individual pixels to match.

vegetable said:

Phranc said:

At what ISO would consider using noise reduction like Deep Prime for either of these cameras? Thanks!

Click to expand...

Deep Prime is probably not the right kind of noise reduction algorithm. It's not their fault, it's Canon's fault. Take a look at the RAW noise in Canon cameras, it's not real. Either by virtue of some flawed de-bayering algorithm, or by intentional manipulation, Canon destroys the pure sensor readout and noise patterns. Of course, all cameras do this now which is why customers have been locked into these upgrade cycles. Not that it's all bad of course, getting better more capable gear is great, in theory. The assumption we all make is it really is better and more capable. But since they manipulate the sensor noise after the fact, they could be leading us down any one of their chosen rabbit holes.

Assuming noise is even a real phenomenon, it may not be, it would be very easy to remove using the same tech they're using for the AF. Machine learning. The key thing is that noise is not random, it's generally related to the input. Because of that, noise follows a pattern related to the image, machine learning could easily remove the noise by being trained on 10's of thousands of noisy images. The thought is, the algorithm could then see how the noise tends to follow the shape of the image and simply correct the individual pixels to match.

Click to expand...

It's difficult to reply to this while remaining polite.

Not sure what to suggest for a technical discussion of noise in digital cameras, but here's a Wikipedia link.

I doubt that most cameras reach the shot noise limit at high ISO settings. I'd guess that electronic and quantization noises dominate.

Post processing of a single frame cannot increase its accuracy without adding a priori information. Noise reduction at the moment is about making a picture look good. The "plastic" remark is about excessive noise reduction making people look like mannequins. It's sometimes seen when old film movies are scanned and undergo noise reduction for release as UHD (sometimes called 4k) digital movies.

So much depends on how much cropping there will be.

With a high degree of cropping, noise is pretty apparent at ISO 800 and over.

It really does not matter what the camera is

Deep Prime does both noise reduction and lens corrections. It is tailored to work at every ISO level, taking the specific characteristics of the camera, lens, and ISO level into account. I use it for every shot, including at base ISO. It doesn't produce 'plastic' looking images, unlike some other NR programs. In fact, it increases and enhances detail. If you use the version that's included in the full Photolab (now on version 5), you can also fine-tune the effects of both the NR and the lens corrections. The idea that noise reduction has to produce 'plastic' images comes from the days when most noise reduction programs simply smeared details. That's no longer the case with Deep Prime (and maybe some of the other recent AI-driven NR programs). The sharpening and detail-enhancing of Deep Prime is so good that I almost never have to add any more sharpening. Most of my images now are just processed in PL5 (which has many other great tools built-in, and can be combined with the film emulations of Film Pack, and another DXO distortion and perspective correction program called "ViewPoint"). When I do want to do stuff in Lightroom or Photoshop as well, I simply export the image from Photolab directly to LR as a DNG. Then I zero out all the sliders in LR, because all the Photolab settings are already applied.

TukTuk said:

Phranc said:

At what ISO would consider using noise reduction like Deep Prime for either of these cameras? Thanks!

Click to expand...

I 'd not consider making a plastic fantastic at all, but applying NR depends FIRST OF ALL not on ISO but on exposure... properly exposed "high" nominal ISO requires less NR than badly underexposed "low" nominal ISO (or vice versa) - more so when camera's sensor actually delivers less read-related noise at higher ISOs (like dual-gain sensor or older Canon sensors with ADC off chip... excluding cases with intermediate ISOs)

Click to expand...

I agree, but it also seems to me that the amount of noise in dark areas on an EOS R5 depends upon camera temperature. More noise is visible if one enables peripheral illumination correction or some of the other features that would result in more noise in darker areas.

Looking at the noise reduction value chosen by the camera in DPP shows some of this variability.

Sometimes the camera seems to mistake high contrast bird feathers for noise.

Phranc said:

At what ISO would consider using noise reduction like Deep Prime for either of these cameras? Thanks!

Click to expand...

I apply Deep Prime to all my images (from my RP and now my R6). I have never noticed any plastic look, nor any reason to turn it off for low ISO.

I run it at the default settings up until about 3200-6400 ISO, from there it gets minor increases as ISO increases, up to about 60-70% at 12800-25600. Shoot I even get usable images at 51200 ISO without maxing out Deep Prime (anything over 80% starts to look pretty bad.

My primary use case is capturing important moments, as I dont have the best memory to begin with. So ill take a technically imperfect final result over not having the shot at all (hence 51200 ISO shots).

I use Deep Prime from DXO. It is really a case-by-case decision. It is the last edit I apply. I zoom in to 400% to get a sense of how much noise there is and the trade-off between noise reduction and sharpness. I've had shots at ISO 6400 and above to which I did not add any noise reduction and others at 100 to which I did. An example of the latter is a 30 second night shot taken recently at ISO 100 to which I added Deep Prime AI.

Alastair Norcross said:

Deep Prime does both noise reduction and lens corrections. It is tailored to work at every ISO level, taking the specific characteristics of the camera, lens, and ISO level into account. I use it for every shot, including at base ISO. It doesn't produce 'plastic' looking images, unlike some other NR programs. In fact, it increases and enhances detail. If you use the version that's included in the full Photolab (now on version 5), you can also fine-tune the effects of both the NR and the lens corrections. The idea that noise reduction has to produce 'plastic' images comes from the days when most noise reduction programs simply smeared details. That's no longer the case with Deep Prime (and maybe some of the other recent AI-driven NR programs). The sharpening and detail-enhancing of Deep Prime is so good that I almost never have to add any more sharpening. Most of my images now are just processed in PL5 (which has many other great tools built-in, and can be combined with the film emulations of Film Pack, and another DXO distortion and perspective correction program called "ViewPoint").

Click to expand...

+1 I put that sh!t on Everything! :-D

R2

Ditto. My workflow starts with PL5 and DeepPrime on everything, although I prefer ACR and Photoshop for most of the rest.

vegetable said:

Phranc said:

At what ISO would consider using noise reduction like Deep Prime for either of these cameras? Thanks!

Click to expand...

Deep Prime is probably not the right kind of noise reduction algorithm. It's not their fault, it's Canon's fault. Take a look at the RAW noise in Canon cameras, it's not real. Either by virtue of some flawed de-bayering algorithm, or by intentional manipulation, Canon destroys the pure sensor readout and noise patterns. Of course, all cameras do this now which is why customers have been locked into these upgrade cycles. Not that it's all bad of course, getting better more capable gear is great, in theory. The assumption we all make is it really is better and more capable. But since they manipulate the sensor noise after the fact, they could be leading us down any one of their chosen rabbit holes.

Assuming noise is even a real phenomenon, it may not be, it would be very easy to remove using the same tech they're using for the AF. Machine learning. The key thing is that noise is not random, it's generally related to the input. Because of that, noise follows a pattern related to the image, machine learning could easily remove the noise by being trained on 10's of thousands of noisy images. The thought is, the algorithm could then see how the noise tends to follow the shape of the image and simply correct the individual pixels to match.

Click to expand...

Hey, after all it's Canon's code used for the software and firmware in the R5 after all. Can't blame them. Prime does a very good job indeed. But with the latest version of Topaz I've become a believer. I wasn't at first but with their latest update, if you take control (which it allows in either RAW or Jpeg you can achieve perfection in virtual elimination of grain but then recapture unreal levels of detail with following it up with the Sharpen AI. I can't say enough how these programs have transformed my hatred of grain and noise generated by high ISO shooting. Now I'm fearless. LOL