Good question.
I shoot raw only, not raw+JPEG. I'd expect this setting to have no effect if it didn't apply to raw files, but I get the slower processing even though no separate JPEG files are being created.
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
OM-1 - Low ISO Processing
- Thread starter iwaddo
- Start date
I shoot raw only, not raw+JPEG. I'd expect this setting to have no effect if it didn't apply to raw files, but I get the slower processing even though no separate JPEG files are being created.
Camera JPEGs.
Indeed. Please see my reply to drj3 earlier in this thread.
There I referenced another post where I was speaking in detail on buffer clearance times:
https://www.dpreview.com/forums/post/67510965
I seem to recall there was another post where I gave detailed information on that subject but I can’t seem to find it at the moment.
knickerhawk
Veteran Member
This feature was first introduced in the EM1x and then as a firmware upgrade to the EM1ii. It would have been kind of a big deal (both negative and positive potentially) if it did something at the raw level, but it doesn't. A second stage of (JPEG image processing) noise reduction is added to the "Detail Priority" mode, which adds processing overhead. Hence that mode results in the slower clearing of the cache.
See here: https://www.dpreview.com/news/52974...ures-from-olympus-s-e-m1x-to-the-e-m1-mark-ii
Last edited:
That makes sense, even if still applying that extra JPEG image processing when exclusively shooting raw really doesn't. I guess there's a JPEG preview embedded in the raw files, but I don't think many people would want to compromise speed for the sake of that looking less noisy.
It would be nice if there was anything in camera, or in the OM-1 manual, to indicate that this only applies to JPEG files.
Assuming that this is the cause of slower buffer clearing when shooting above ISO 1600, I wish it was possible to turn it off even when shooting at high-ISO. In fact, at high-ISO I'm more likely to want a raw file that I can run through denoising software.
drj3
Forum Pro
Has anyone seen any Olympus/OM Systems statement that it only affects jpegs. There is one from the DPR E-M1X news, but if you read the Olympus statement they quote it does not say jpeg. Neither did the engineers that Imaging Resource talked to.
Can we believe Googles AI
does detail priority mode affect only jpeg images with the Olympus cameras
"No, the Detail Priority mode on Olympus cameras is not exclusive to JPEG images. While Detail Priority can enhance JPEG images, it also impacts RAW (ORF) files when shooting with Olympus cameras. The mode works by optimizing the camera's processing to extract maximum detail from the sensor data, which is then applied to both RAW and JPEG files."
Google seems to have snagged that from Thomas Eisl, who has sometimes spouted nonsense in the past, so that is not a reliable source.
drj3
Forum Pro
My trust of most AI reports is close to zero, so I agree. However, I am not sure that the statements that is only affects jpegs has much better support, which is why I ask if there is a source of better information. I do not doubt there is an effect on the jpeg images, but have seen no good source indicating that it does not affect RAW images. It does affect the buffer capacity for RAW only images.
Given the relatively subjective evaluation of small differences in detail and the real focus variability of cameras/lenses it would take a very large number of carefully controlled trials with blind detail evaluations to determine if and how much it might affect RAW images. Choosing an appropriate target would also be necessary.
One would have to treat each burst as a single trial. For my tests of focus accuracy I used from 50-80 trials per condition. Having to evaluate all the images in each burst for a large number of independent bursts is not something I am likely to do, but I would like to know the answer.
SrMi
Veteran Member
Have you checked? When searching, the AI answer gives links to sources it uses,
My search for "Does Olympus detail priority affect only jpeg" returns four references, none of which is from Thomas Eisl. Also, none of them seem to support the claim that raw files are affected.
On the other hand, Reinhard Wagner, a reputable resource, claims that it affects raw files.
I'll believe it only when I see a difference in the raw data.
I have found AI answers to be very confident but unreliable.
Actually, I did. Otherwise I wouldn't have posted.
The Google AI "responses" for your exact query are typical LLM garbage. If you check the links it references to your exact query, none of them are even about the setting in question, so they are irrelevant.
Likewise, the results to the way drj3 posed the question to Google.
Since "detail priority mode" is not the name of the setting, but rather it is "Low ISO Processing" I phrased the query in a way more likely to hit relevant pages:
om1 does "low iso processing" affect raw
The quotes are necessary.
My query actually hits relevant documents, that actually mention the "Low ISO Processing" setting on the cameras in question. Despite that, as I said, they are from Thomas Eisl and are unreliable.
It's worth mentioning that this question was discussed years ago when the feature came out on the EM1X:
https://www.dpreview.com/forums/thread/4494062
And then later again here:
https://www.dpreview.com/forums/thread/4565553
And then yet again here:
https://www.dpreview.com/forums/post/66033743
Nowhere is there an actual statement from Olympus one way or the other.
Someone in the first thread claims to demonstrate that it does not apply to raw, but the visual evidence he provides is, to me, inconclusive. You can find people claiming either answer is the correct one.
So, 'tis clear as is the summer sun.
SrMi
Veteran Member
I see with your query Thomas Eisl's "contribution."
We agree on the lack of reliability of AI answers.
drj3
Forum Pro
This is a situation where the difference is likely to be relatively small compared to the focus variability of the specific camera/lens, which implies that a large number of trials would be required to reach any reasonable level of confidence in the result.
The first problem would be selecting a target that is sufficiently sensitive to detail differences to demonstrate any change in detail. A double blind study would not be necessary assuming the treatment order is random and process is basically automated, but the assignment of detail level (dependent variable) would need to be blind to the condition evaluated.
Each image of each burst series would need to be evaluated and then an overall detail score assigned to that burst. In order to do that one would initially need to create a series of comparison images with different levels of detail for comparison to the burst images to assign detail scores.
Maybe if I am snowed in for a couple of weeks next winter I will check it out. However, if the weather is not that bad, then I will probably be out actually photographing wildlife.
I agree completely, drj3.
I previously explained why my advice was “set it to detail priority and don’t worry about it” for most cases at least.
I tested quite a while back, and there’s a small quality implication, but why take the hit unless you really need to minimize buffer clearance times…
Since I’ve already made that determination, it’s not worth it to me to spend the time necessary to comprehensively prove everything about exactly what it does under what circumstances.
I previously explained why my advice was “set it to detail priority and don’t worry about it” for most cases at least.
I tested quite a while back, and there’s a small quality implication, but why take the hit unless you really need to minimize buffer clearance times…
Since I’ve already made that determination, it’s not worth it to me to spend the time necessary to comprehensively prove everything about exactly what it does under what circumstances.
knickerhawk
Veteran Member
I just shot a burst of darkframes (lens cap on at high shutter speed). Examining the JPEGs in Photoshop with a severe exposure push applied, it's easy to see a difference in the noise levels (size and distribution) between JPEGs from the two modes:
SOOC JPEGS. +20 Exposure push to both halfs applied in PS. Left=Drive Priority; Right=Detail Priority
Based on what DPReview had written and the lack of any previous discussions of noted differences in the two modes generated from raws processed outside of the camera's JPEG engine, I expected to see no differences in the raws. I was wrong. With some pretty severe (but equally applied) manipulations in ACR, there is a faintly visible difference in the amount and pattern of the distribution of noise between the two modes. (The pattern is caused by the otherwise generally randomly distributed noise being pushed and pulled by the built-in lens distortion correction applied by ACR:
Raw converted in ACR with same adjustments applied to both. Left=Drive Priority; RIght=Detail Priority
The differences above are very subtle but visible upon close examination. Likewise fast Fourier Transforms (FFTs) performed on the converted raws also show that there's different amounts of what I presume is noise reduction applied (more appears to be applied to the Detail Priority rendering).
It's difficult to avoid the conclusion that something's going on in the raws and not just the JPEGs. This quick test tells us little about what differences (if any) could be seen in real world photos generated outside of the camera's JPEG engine. It's worth some more testing, but I'm working on something else at the moment and might not be able to contribute much for the time being.
The differences between the two raws you show are, at best, subtle. I'm not sure if I see what you think you see in the raw example.
I'm not disputing your conclusion, but I'm curious about something regarding your methodology.
Every raw is going to be different because the noise is going to be different each time; additionally subsequent shots will be noisier than earlier shots as the sensor heats up.
What happens when you compare two "Detail Priority" shots to each other, or two "Drive Priority shots to each other, FFTs and all?
It needs to be established whether the difference you think you see is actually due to changes in the setting, or if similar differences exist shot to shot even with the same setting.
I'm also not sure that the way any NR algorithms react to electrical noise in the sensor (which is what you are capturing) will be the same as how they react to an actual exposed image. An actual image might be more probing. Some low contrast pattern fabric, perhaps.
Last edited:
Wow, I never anticipated such a debate, I much more expected a rather binary answer.
I remembered that I had recordings of the OM Menu course.
In this recorded session it is mentioned that Low ISO Processing affects ISO 800 or lower but it is not mentioned whether it is JPEG, RAW or both.
Thank you all for your help.
I remembered that I had recordings of the OM Menu course.
In this recorded session it is mentioned that Low ISO Processing affects ISO 800 or lower but it is not mentioned whether it is JPEG, RAW or both.
Thank you all for your help.
Last edited:
knickerhawk
Veteran Member
Well, maybe you should dispute my conclusion because I'm now reversing it. (See below.)
With respect to noise, other than some fixed pattern noise which should be generally similar from one blackframe to the next, the only noise present should be read noise (remember, they're blackframes, so no photon shot noise). Read noise is random. If there is no noise reduction or other forms of digital signal processing applied to the raws, we know what the FFTs should look like (i.e., just a uniformly dark image with the possible exception of very thin intersecting middle horizontal and/or vertical lines due to manufacturing-induced edge light falloff). For a nice introduction on what to expect see here.
My bursts were quite short (5 frames in each mode). They were separated by a couple of minutes as well, so heat build-up isn't really a concern.
In my first test, I checked all of the frames. They were uniformly similar within each set, both visually and with respect to the FFTs. The same was true in the second round of tests explained below. Just what you'd generally expect from blackframes generated with consistent settings and conditons.
I agree that shot-to-shot variation is something to be checked, and I DID check for variation. I just forgot to mention this in my prior response.
I'm not aware of any claim ever made for Oly/OM raws having varying levels of NR applied based on image-specific content as opposed to fixed (known) conditions such as ISO. I would be shocked if that's a factor here given the large processing impact it would likely have and the fact that it has never (to my knowledge) been identified as a feature.
Having said all that, I re-ran my tests today. Two things changed. I changed the lens used (previously used the Panny 25mm f/1.4 and today used the Oly 12-40mm f/2.8 set to 12mm). This time, in addition to processing the raws in ACR, I also processed the raws in RawTherapee so that I could turn off the auto lens distortion correction that ACR applies and see how that difference showed up in the FFTs.
The JPEG results (which, of course, weren't affected by the change to RawTherapee) continue to consistently display the same visible differences in noise amount and size. In this new test, the ACR-generated raw results continued to show the distinct signature caused by the lens distortion correction. However, the size of that "signature" did not vary between the two modes like it did when I used the Panny lens. I was actually expecting a larger "signature" for the 12mm lens due to the greater distortion correction required, but that wasn't the case. I also have no explanation for why I saw a difference in the size of the signature between the two modes in the first test but not in the second.
When the raws were processed in RawTherapee, the distinct FFT "signature" was present when lens distortion correction was turned on and completely gone when it was turned off. (This proves that the distinct signature visible in the FFTs was due to lens correction applied by the raw converters and not something irreversibly built into the raw data itself.)
With the lens distortion correction turned off in RawTherapee, the FFTs from both modes were very clean looking and indistinguishable from each other. There was not any obvious sign of raw noise reduction either (in either mode). I also averaged the 5 shots from each mode and FFT'd them and compared these averages. Again, there was no meaningful difference at all in the FFTs between the two modes, either visibly or when compared in PS using the "Difference" blending mode.
I can't account for why there was a difference in the raw FFTs for the Panny lens and none for the Oly lens. That's an interesting anomaly, but I'm more inclined now to agree with the DPR article's reference to the difference only being in the SOOC JPEGs and NOT in the raws themselves. Additional testing to clear up the differences in my two rounds of testing would be ideal, but I'm not particularly motivated to spend more time on this. Sorry for any confusion I've contributed to the discussion.
No, there's also dark current noise, which is quite distinct from read noise. It can consist of both fixed pattern and temporal noise. It's an inescapable property of a real-world sensor system. It also is impacted by temperature (both ambient, which varies, and operation-induced heat (which is why I commented about the sensor heating up for subsequent shots). This is why imaging sensors in critical or demanding environments (like astronomy or astrophotography, where images are mostly black) are actively cooled. By leaving the lens cap on, you're reducing the signal to zero, so noise dominates. You can read more about it here .
Firstly, there are lots of things that are true that Oly has not publicly stated.
But more importantly, you seem to think I'm talking about some sort of content recognition, but I'm not. Many noise reduction and compression algorithms are adaptive; they perform quantization and frequency analysis on the data to optimize performance. Even rudimentary frequency analysis of an image will reveal a photo of a test target or a tree is very different from a photo of noise, because they have structure. We do not know that this sort of thing is not going on inside the camera, and we know it absolutely is happening in the case of generating jpegs. If one mode wants to preserve "details" then it stands to reason it might be imbued with some simple definition of what constitutes a "detail" that distinguished detail from noise.My point before was that when we don't know for sure what's happening inside a system, it's best not to assume that nothing is happening. See relevance of this thought below...
Good catch on your part. Lens Correction was not something I specifically thought of, but it goes back to the principal of not assuming nothing's going on. There was indeed something going on that we didn't realize, that left a reproducible fingerprint in the data, and you found it.
Again, good catch. And I feel vindicated in my skepticism.
Even though the hypothesis from your images wasn't proven, it was a good discussion.Hi
I’ve just received a reply from OM-Systems
’I can confirm that the LOW ISO PROCESSING options will be active for RAW and JPEG if those formats are actively set too.’
So my RAW images at 800 ISO or lower will be affected by choosing either Drive or Detail Priority.
Thank you for your help.
I’ve just received a reply from OM-Systems
’I can confirm that the LOW ISO PROCESSING options will be active for RAW and JPEG if those formats are actively set too.’
So my RAW images at 800 ISO or lower will be affected by choosing either Drive or Detail Priority.
Thank you for your help.
knickerhawk
Veteran Member
And now, based on the response Iwaddo received from OM System, I should reverse my reversal. I'm feeling dizzy.
On a slightly more serious note, I'm a little skeptical of reading much into the response provided by OM System. These things have a tendency to lose validity in translation. We've seen "official" and supposedly reliable sources get things mixed up. I will treat the response as "interesting" but not definitive until an actual detailed explaination of what's actually happening under the covers is offered up. Oly/OM has treated this "feature" as an enigma wrapped in a riddle from the get-go. Why?
Also, I want to make very clear, if my prior posts fail to adequately convey it, that my testing is limited and shouldn't be relied on (as if my flip-flopping speculations haven't made that clear already). Generally speaking, though, I'm not aware of a better and more reliable starting point for detecting and evaluating digital signal processing in raw files than looking at blackframe FFTs. Of course, evaluating blackframe FFTs have limitations and particularities of the feature in question here (specifically, the setting only applying to bursts) add time-consuming complexity. Nevertheless, based on the limited and fairly casual testing I've done and compared to the obvious differences in the JPEGs, I'm just not seeing much to get worked up about with respect to raws. The lens correction anomaly is intriguing but not enough to interest me to pursue this further. Perhaps you'll pick up the scent trail and do your own???
I concede that what I wrote oversimplified by not acknowledging contributions from DSNU and DCNU. However, the impact of DCNU in particular on the testing I did is surely next to nothing. We're talking about a five shot burst at a shutter speed of 1/4000 (taken in my cool dark basement utility room). Unlike the astrophotography use case you refer to where thermal noise is a significant factor, the impact here when (for instance) comparing the first and last shots in a burst or between the separate bursts is really de minimis. Besides, I already accounted for it by looking at all of the frames in both bursts and FFTing all of them and also averaging each set and looking at those FFTs. The highest difference in the raw standard deviations between the first shots in each burst and the last shots was approx. 0.01 in the red and blue channels and 0.02 in the average of the green channels. Again, feel free to do your own testing and calculate the statistical significance of the DCNU impact on the testing. I'm just not seeing how this is an obstacle or unavoidable confounding factor in my (admittedly casual and preliminary) speculations as between the bursts. Are you saying we should expect different amounts of thermal noise in one mode vs the other? I just don't see it being a factor, at least when comparing like to like (i.e., corresponding frames from each burst or averages of all of them).
Sure, it's going on with JPEGs, but please point me to evidence of the kind of image-specific adaptive DSP (as opposed to other adaptive DSP based on things like ISO) that are currently being utilized at the raw level in any of the cameras that show up in DPReview. (I'm not interested in responses about the compressed raw options that some cameras have. They aren't relevant here.)Firstly, there are lots of things that are true that Oly has not publicly stated.
I fear that the conversation is swerving into the wildly speculative sphere here. Is it possible that Oly/OM developed some super-secret, fast and smart adaptive DSP that somehow improves raw "detail" while also evading detection in blackframe FFTs? Well, yes I suppose it's possible. Is it possible that Oly/OM's marketing department and technical documentation team chose not to tout this feature? I suppose that's possible. Is it possible that in the six-or-so years it's been available none of us ordinary Oly/OM users have picked up on how this feature visibly impacts IQ rendered from raws? Again, I suppose it's possible. Is it possible DPReview staff got it wrong when it wrote: A new 'Detail Priority' mode reduces noise (at the expense of burst speed) at low ISOs in JPEGs. Well, it wouldn't be the first time DPR was wrong!
See also: Occam's Razor...
I'll just add one last thought before exiting this discussion. I bought a Nikon D300 when it came out. One of the touted features was the option to switch between 12-bit and 14-bit raw capture. Somewhat unlike the situation here, the visible IQ superiority of the 14-bit raws was hotly and frequently debated early on, but the consensus around that conclusion pretty quickly solidified. There was just too much evidence supporting the claim. It turned out that most of the reason for the better IQ was not due to the greater bit depth but, rather, because the read-out speed was much slower and, therefore, more accurate (less noisy). Something along the same lines could be going on here based on the differences in burst speeds between the two modes. In other words, there might be a [very] modest SNR improvement in the raws generated from Details Priority mode (hence the references to improved raw IQ). There might ALSO be some DSP voodoo differences in the two modes at the JPEG level - e.g., a bit of extra NR splash to SOOC JPEGs generated in the Details Priority mode (hence the claims promoted by DPR and elsewhere). To me, this is one way of reconciling the seemingly at-odds information and evidence we're grappling with here.Good catch on your part. Lens Correction was not something I specifically thought of, but it goes back to the principal of not assuming nothing's going on. There was indeed something going on that we didn't realize, that left a reproducible fingerprint in the data, and you found it.
Again, good catch. And I feel vindicated in my skepticism.
I could be wrong...it's been known to happen from time to time. ;-)
Last edited:
Similar threads
Keyboard shortcuts
- f
- Forum