Well, that made my brain hurt but well worth it. I wouldn't gave guessed I will likely use the mid resolution when my Q3 finally arrives. Thanks so mush for sharing this work!!
Thank you Ted. I did this work for me to better understand the Leica pixel binning topic and then I thought why not sharing my thoughts with everyone on my channel :)
@@mathphotographer So helpful, but now another equation if I may. While I wait for my Q3, I ordered a pre-owned Q2M which I'm very excited about. Do you have any direct knowledge or even a guess about ISO-Invarience for the Q2/Q2M? Is it a stretch to think that the mid-resolution would offer better performance as you found on the Q3? I understand that you may not be able to extrapolate but any recommendation would be most appreciated, Thx!! So excited to enter the Leica world!!
oh man...Du bist geil. Honestly, I was torn between buying the Leica Q2 Monochrom and the Leica Q3 because I want to take black and white photos in low light conditions. After your detailed and scientific explanations, I will definitely buy the Leica Q3.
WOW. I love this. No bullshit, no speculation, no subjective opinion. Straight up measurement and definitive testing and conclusion. If only RU-vid was filled with people dedicated to facts and figures as yourself it would be a much more educational and valuable place. Thank you.
Holy cow! How I like when you get nerdy and geeky :D my head exploded, but it was worth it. Many great photographic channels on RU-vid, yours is way ahead of them. But what can I say, my Q2M is already at the Leica store and I am waiting for my Q3...
Amazing analysis! Going from L to M gets you the classical 3dB per 2x oversampling. Going to S suggests you may have started to get into under sampling regime. And I know the person in the Kodak Man reproduction. He is now an IBM technical fellow!
Thank you indeed. This is like radio Veronica in the 60's or 70's: u vraagt wij draaien (all old belgians or dutch will know this). If all would be so responsive life would be a lot easier. Thanks again.
I just recently developed that toolkit I showed in the video and have only applied it to the Q3 at this point in time. Of course I will also apply it to other cameras now like the M11. My guess is that the result would be the same for the M11 since the sensor of the Q3 is likely very close to the sensor of the M11.
Grüezi! I might have skipped this info, but to which resolution were all images (60, 36 and 18 mpix) downsized before running all the scripts? If the were saved to lossy format, how did you control quality?
No down-sizing, out of camera RAW and then in some parts running the tests over the RAW files and in some tests running it over the JPG images after export from RAW into JPG in Lightroom. Since I compare only images with the same resolution, so e.g. M-DNG high ISO with M-DNG low ISO, the resolutions always match. No need to change image size.
@@mathphotographer Hmm then I don't get what you're showing at 20:30. You're clearly comparing M-DNG and S-DNG to L-DNG, yet for each one you're using different resolution. This test does not prove, that in-camera pixel "binning" improve SNR, it does not rule out that it may be an effect of simple downsizing. Would be good to compare apples to apples and downsize all files to the same resolution. And it's worth to use lossless image format for any scientific comparisons like TIFF or PNG instead of JPEG.
@@paprycjusz7 The difference images I calculate are within a resolution category, so pairs in L-DNG, M-DNG and S-DNG. But then PSNR and SSIM are compared across the different formats in the same way as Leica is suggesting. Otherwise the exercise would be pointless :)
@@mathphotographer I see your point and understand why you did it this way. What I'm trying to understand, let's say I want to take a photo that will be printed on A4 at 300 dpi (~3500 x ~2500 px), I want to have as high DR as possible. Does it make a difference to shoot in L, M or S? Am I going to get better DR using different formats or not?
what a fantastic channel! have you also measured the actual sharpness? the last test I've seen made a LR setting of +15 with 0 details the optimal non-oversharpened sharpeness. additionally, have you done some tests with the sharpness levels before ai denoising in LR? i.e. are the images better if you set sharpness to 0 and then denoise?
I always enjoy your reviews, particularly your rigorous quantitative discussion of many of the issues. With respect to the Q cameras, I have a question regarding the Summilux lens used on these cameras. How does this lens compare optically with 28 mm Summilux for the M system cameras? I have also observed that the M11 and the Q3 handle white balance very differently. The M11 pictures appear to be much cooler than the Q3 pictures and the color temperature shown in Lightroom is generally much lower for the M11 than the Q3. Have you observed this? Finally, one question regarding this video. How can you check for 14 stops of dynamic range with an 8-bit Jpg file? Thanks.
Generally comparable but actually on the Q3 fine-tuned for the special situation that the prime lens is fixed mounted on the camera body and serves just this one particular camera as a lens. And the Q3 has a lens setting for close-ups and macros ... clearly nice to have :)
Hi, just get the Q3. May I know M-DNG / S-DNG get better noise performance or using the Lightroom AI Denoise on L-DNG and reduce the size should be better? Thanks
Many thanks, Jason. I got the wireless charging station for the Q3 from Leica Zurich yesterday and the grip will come soon although not yet available in Switzerland.
In a previous video on M11 pixel binning you concluded that shooting at 60MP (and resampling to 36MP in post-processing) would provide superior resolution and SNR compared to shooting at a pixel-binned 36MP. That conclusion seems at odds with the conclusion in this video since the M11 BSI sensor is similar to the Q3 BSI sensor. Am I missing something?
I didn’t see any resampling of the L-DNG down to M-DNG in this work, but it would be very interesting to find out using the same method shown here. Also I’m not sure how the so-called pixel binning differs from a downward resampling. It may be the same thing - only Leica knows.
Your observation is correct. I just recently developed this new testing routine in the Q3 video and have not applied it yet to the M11. Its kind of the same sensor as in the Q3 so I expect similar results for the M11. As I said in the video, its hard to "see with naked eyes" the effect of pixel binning, you really need to run some tests with hard-coded metrics to see whether pixel binning is effective or not. So far I've done this only recently for the Q3 but will run similar tests for the M11 in the next weeks, promise :)
Good point, I believe S-DNG is there mainly for file size reasons although, as you see in the video and the Leica data sheet, still superior when it comes to dynamic range. For me M-DNG works best: improvement in dynamic range and noise behaviour and with 36MP kind of an M10R-resolution, really good in my opinion :) I was always happy with the 40MP of my M10R.
You are one impressive scientific photographer. Being a RF hardware engineer working at Apple in Cupertino, USA, I perfectly relate to the DB, MSE and SNR, but I also learned something new: Structural Similarity Index !!!. I also Lear Ed that you can write python scripts. Incredible talent!!. So for this Q3 sensor construction, the sweet spot is 36MP, and I assume that for a different density either a different pixel size (like the one in Hasselblad’s X2D) there would be a different pixel binning sweet spot (were it to be offered). Many thanks for the both the educational and entertainment value you provide in your videos.
Now I see where your ”mathphotographer” channel name comes. This was really enjoyable to whatch and informative at the same time, thank you for your effort!
Profoundly awesome educational video. I finally understand why Leica does the three choices for resolution. I’m wondering if the next get SL2, the SL3 will have the LDNG, MDNG and SDNG, which will mitigate the need for a second camera (SL2 vs the SL2-S) with lower megapixels for low light shooting and video.
Fully agree, the SL3 which will likely come this year will for sure have L-DNG, M-DNG and S-DNG. Leica has developed here some nice tech and we will see it in more cameras going forward.
Hi, I also have the Q3 (and an M11) and this is a great video! Quick question though, at ~20:22 you show that the ISO SNR has improved from ~30 dB to 33dB and that you say that's a 10% improvement. However, since SNR is measured in dB and it's a logarithmic scale, so 33dB is twice as much as 30dB wouldn't it be more precise to say it's a x2 improvement?
Hi - this is a great video, thanks for all the related production efforts, and thanks for sharing! I struggle though with the concept of 'ISO invariance' - can you help me here? - I assume in your examples you shot (nearly) identical pics with the same exposure and aperture; just varying ISO settings - I assume you shot RAW pictures - The sensor does not know about the concept of ISO. It records the amount of light per pixel, as per exposure / aperture settings - The ISO is added as a metadata value to the RAW file - The actual raw data is always the same, irrespective of the ISO settings in the metadata (that's different from the analog film world, where the "hardware" / the characteristic of different ISO films differ and can't "record the same pic") - When opening and displaying a pic e.g. in Lightroom, Lightroom takes the raw data of the pic, and uses the ISO metadata value to make the picture lighter / darker (increases / decreases the exposure behind the scenes) - So what you do if you push one of the pictures is you offset what Lightroom just was doing. Now you compare two pics with nearly identical raw data, displayed with the same ISO ratio - assuming that the algorithm for initial displaing the DNG file and later changing exposure is the same (for working with RAW data, that may be a fair assumtion), there should be no difference in the results. So what you've kind of proven is that x = y + 3 - 3, i.e. that x (pic 1) equaly y (pic 2) when changing (via ISO setting) exposure, and then offsetting this change manually. Right? So in that sense, wouldn't the "ISO variance" at best be a shiny marketing concept? Would love to get your views, thanks!
Excellent presentation. I have been shooting L-DNG's and been enjoying cropping in and the expanded tonal range of adjustments in Lightroom. The Denoise feature in Lightroom is remarkable. As you point out the M-DNG will get the Q3 to15 stops. Will try M-DNG's and see if the extra stop is worth the loss of resolution, or smaller file size is beneficial. Thank you keep up the good work.
In another video you should upscale a 15MP photograph from the 60MP sensor and compare it to the same scene shot at 60MP. There you will discover that there is only a very minor difference, some extra sharpness in the green channel. That is because each pixel only measures 1 colour (RBGG per Bayer micro-lenses) and the rest are interpolated. This is why the Leica monochrome cameras are such sharpness beasts!
Probably worth mentioning that this 60MP sensor is most likely made by Sony, possibly with tweaked micro lenses. The performance and specs are otherwise identical.
This is the best evidence based Video i found so far for judging objectively the capabilities of the Q3 . As I am in the decision to invest in Q3 or Q2 monochrom - I wonder if there is a significant difference left when it comes to low light performance in b/w when taking post production AI noise reduction into account. At least fore a non professional Shooter not heading for the last percent I tend to think the trade off for more flexibility with Q3 (in respect of the colour option by still having a near to Q2 monochrom low light performance ) is acceptable. - I don’t if you made Investigations on this question exactly. Could you give an answer on this?
The Q3 is superior to the Q2, newer sensor, new algos implemented. But the sensor of the Q2 is not far from the Q3 although the Q2 has a lower resolution and does not allow for M-DNG and S-DNG. I played with a couple of sample images of the Q2 and applied the same toolkit as in the video, the results are not as strong as on the Q3 but also not bad. I think if you have a Q2 and are happy with it, there is no urgent need to upgrade to the Q3. If you want the best of the best, then the Q3 is for you :)
@@mathphotographer thank you for the reply! its more of the low light colour noise and blowing out of highlights that concerns me with the q2. The M-DNG in the q3 seems like the perfect balance and a considerable jump in both of those aspects over the q2 with close to the same resolution from what I've gathered. the question is, is it worth (almost) twice the price I paid for my q2? that's where im not so sure.
It would be interesting if you could make a video about the methods of reducing the amount of pixels in a camera. I think that Leica does NOT use PIXEL BINNING as you called it in your video. This might be a detail as the video covers fully other topics. But as a matter of interest it would be great to fully understand what pixel binning does compared to other methods.
Another outstanding, beautifully illustrated and expertly explained video. Leica ought to give you a consultancy contract, if only for teaching. I would love to see you analyse the Q2 sensor in the same way, if you would have time. It would enable us to objectively and scientifically understand the differences better. By the way, with an ISO of 3200 in both my Q2 and SL-2 with APO Summicron 75mm 2.0, DxO PureRAW 3.0 allons me to recover all details I need, including the very dark areas, from images taken of ocular gene therapy interventions in the operating room. Thank you for your videos, simply superb, Bart
Many thanks, Bart, very nice feedback. I will apply the toolkit I now have to other cameras for sure, including the M11 and M11 Mono and likely also the Q2.
A masterclass! Your tests were done with binning. Do you think the same applies when digital zooming is used at for example 35 and 50 mm%. I don’t have the Q3 and perhaps binning is not possible with digital zoom.
Excellent question, and thanks for your kind feedback. I am not sure whether digital zoom will behave in consistency with native focal length where the full sensor is deployed. Let me look into it and come back.
I loved this test but I have a question from a more image oriented photographer. I would really like to see in the images, the highlights behavior in L an M size, and in the different iso. Because one thing is the DR related to noise. But other very different thing is the Rolloff of an image, that specific latitud in the highlights before the critical clippling of the whites. I had the Q2 and the camera even when in every test appears to have a very good dynamic range, no matter how you expose, there is always a clipped highlight in the frame. Don't know the reason, but from an aesthetic point of view for me is very unpleasing. The other day I had the opportunity to play for some minutes with a Q3 and I bring the files to my computer, What I have noticed immediately is that the files has a lot less noise in the shadow and when you recover shadows, but the highlights still has this contrasty behavior, the good thing is now you can underexpose by 2 to 3 stops confidently and will have good shadows. The problem is that you are going to be looking at a very dark viewfinder, and just trusting in the histogram or your knowledge about metering. Is there any possibility that you make video talking and showing examples specifically for the highlights? I would really love to see for example how you can recover highlights in the iso 200 image underexposed vs the 3200 image correctly exposed. Because visually that kind of exercises could have a different result. May I ask you why did you compare iso 200 overexposed with 3200 and not iso 50? I understand that iso 50 is the highest dynamic range. In the Q2 there was myth about the iso being less contrasty in the whites. I know astro photographers are more concerned about shadows and noise when we talk about dynamic range, but as a daylight photographer and cinematographer I am more concerned about highlights and clipped skies. I loved my nikon d850 and I really hated my Q2. I would really love to love the q3 this time, but I don´t really trust their highlights. Visually speaking, do you have any example comparing L and M highlight recovering? Thank you again for you beautiful and scientific work.
Thank you, great analysis. I understand the math, but intuitively it seems that the higher amount of pixel binning in S-DNG should result in at least as good an SNR value as M-DNG, even if it didn't further improve over M-DNG. However it is actually a lower SNR value than M-DNG and from a perspective of the physics of the problem, it is not intuitively clear to me why this should be the case. Any thoughts? Thank you again for all of your tremendous content.
I'm confused at why the M-DNG pixel-binning has better results than the S-DNG, which should allow twice as many pixels to "bin"? Per your conclusions at 40:40 in the video. Looks like there would never be a good reason to use S-DNG.
Please correct me if I am wrong, but my understanding is that you do all considerations on pixel level, not on equalised output size, are you? On pixel level it is obvious that averaging several pixels will reduce noise and therefore increase dynamic range (and it can be done in the same afterwards in during raw conversion or after raw conversion), but after equalising the output size (on the monitor or for print), there is no such advantage.
Wow, what a fantastic, educational video! Many thanks for taking the time and effort to educate so precisely and clearly. The new Light Room Classic noise reduction is very impressive indeed. I wonder if you compared the L-DNG at 25,00 ISO vs M- DNG in your detailed train station composition, do you think would the extra resolution and detail of L-DNG would allow the adobe NR algorithm to work slightly better (more data points in fine detail) , despite the slightly better DR of the M-DNG file, or maybe about the same? Also, if 0.5 seconds is about the slowest shutter you can hand hold the Q3, what do you think that slowest acceptable shutter speed would be for the Q2, to get a sense of the improvement for the Q3's OIS? Thanks again for all your thoughtful work and clear, scientific explanations.
Many thanks for your nice feedback. I need to try hand-held shooting again with the Q2 to figure out the improvement of OIS ... will do. On your point on noise reduction in LRC, I think the results would be about the same but need to look into it. Great suggestions.
Thank you for the very detailed exmaples!!! Love the videos you are making with Q3. One question, do you know the ISO invariance range of the Q3? is does this ISO range apply in video mode?
Amazing how many youtubers tries all together this new camera which is yet not availlable on the market !! And very surprised that all tell the same story : it's worth to buy this very good camera, ha ha ha!
❤Thank you for a wonderful video I picked up so many more reasons to change to the Q3 for street photography from my M11, where I am getting camera shake at even 180th of a second when shooting from the hip. I had to wait a long time to get to it but it was so worth it, as I also found out along the way I should be shooting M-DNG for social media (my main use), hopefully this will speed up my workflow as the 60MP images were slowing down my computer. Hoping to pick up the Q3 soon and will use the M lenses (those I will keep) on my SL2-S. Bye Bye M11. Great video Sir.
Great content! I got my Q3 a couple days ago and tested it in low light, however the photos I got (5000iso) are very grainy, probably more grainy than photos taken with my M10R (12500iso), and when I convert it to B&W and push the shadow to recover the details, there’s lots of weird white dots appear. Do you have any idea what caused this issue and how to avoid it? For my m10R, I set the maximum iso to 12500 and I don’t mind having a little bit of grain in the photos, however my Q3 photos with high iso are almost unusable as they are too grainy and I can’t push it since there will be white dots.
Can I assume your analysis applies to the Sony A7rV L M and S lossless compression raw files? M lossless compression would be the optimal for dynamic range…
Very interesting and thorough video, well done. It's easy to explain the constant SNR between ISO200 and ISO3200. The saturation level of the sensor stayed the same because your aperture/shutter settings were always the same. A few more comments: - The dynamic range of the sensor does not change - this can be explained theoretically. - Your focus was to analyze the image - The focus should be the sensor, like: 1. Study the noise floor of the sensor and analyze 2. The rated saturation level of the sensor, ISO range, and DR 3. This will enable you to explain a lot about the camera's performance characteristics - ISO100000 means nearly 10 stops of image signal amplification. That means the sensor received almost no reflected light which could mean you are amplifying nothing or at best the native noise floor of the sensor... See the point. Unfortunately, your study consists mostly of arguments and explanations used in the commercial "size and capture" theory. Perspective or what you are analyzing has a lot to do with what your results look like. I honestly do not think you are purposely presenting a commercial theory. You need to step away from the size and capture theory... What does the size and capture theory promise us? - The bigger one captures more light - The bigger one has less noise - The bigger one has more DR - The bigger one has better image quality - The bigger one.... (you get my point) For example, I wear my biggest shoes when I go walking. I like the math and software approach. It will be amazing to see the results when your focus shifts. See this website and the interesting theoretical discussions and calculations: www.strollswithmydog.com Best Siegfried myolympusomd.blogspot.com .
Clearly a meaningful way to express the ratio. On the flip side of the coin, the way the "signal" is measured, the noise is already contributing to the measurement of the signal. For instance, in the educational section you find the noise already as contributor to the "signal" which in simplified terms in the example is expressed as the mean. Then, in your numerator you would kind of double count the noise, correct? In the video I have of course not "invented" the ratios, they are standard in the industry.
@@mathphotographer I was just indicating that noise is always (already) embedded in the signal, so it seems that the numerator has to include both, but I bow to your expertise. You've created some amazing work here that I will find very useful when I get a Q3.
maybe it’s just my bad eyesight but when you compared the 3200 vs 200(+4 stops LRC), the 200(+4 stops LRC) looks like it has a slight noise/grain compared to the ISO3200 image?? Is it my imagination?
@@ShakeelNaim That's correct in L-DNG when I subjectively look at the side-by-side comparison. But its a small difference, and if you would show me both images without labelling them I would have a hard time determining which one is which.
Very good demonstration. The use of Pixinsight brought back painful memories. This software is not for the fainthearted. My only question is how Leica have done pixel binning so that the MDNG is now 36MP? Usually we used binning of 2X2 or 4X4 pixels to create the super pixel in order to increase the SNR, if we were over sampling the target image. This new sensor of this Q3 seems to be a good one. Thank you.
Many thanks! PixInsight is indeed far from being user friendly :) But quite capable of course. The pixel binning point you raise is an interesting one ... no idea why they follow their own approach here at Leica. In any case, concluding that the Q3 sensor is excellent is spot-on. This is one of the best sensors I ever worked with.
Math, thanks for a very thorough presentation. I have an interest in statistics and you taught me something. :-) The best way to learn any STEM subject, IMHO, is to apply it to something you already know. I wonder, though, if there is further examination to do. For example, you could examine a full 60Mpx image scaled down to 36Mpx and 18Mpx, and perform the same tests. I wonder what the results would be. I predict that you will get more DR than the 36Mpx image. But not by very much. In addition to that, I have further questions about the ISO invariance test. If a scene is shot at both 200 and 3200, the results will be nearly identical, as you have shown. But, is the highlight recovery the same in each instance? If it is, that means that ISO is treated as metadata by the Q3, and the signal is not being amplified. Apparently, the Fuji GFX cameras treat ISO as metadata, just like Red cinema cameras. So, if highlight recovery is possible in the 200 ISO shot, but not in the 3200 ISO shot, that means that shooting at 200 ISO is the better option - although the image might be too dark on the display. I don't know if the Q3 has OVF 'simulation' or not. I use DxO Photo Lab, and I find that the NR is still better than what I have seen here in Lightroom. Not that it matters, because I never use NR too much. I mainly use it to eliminate chrominance noise. There is no point in removing luminance noise, because there is no genuine detail to be recovered. If detail is gone, it's gone.
Many thanks for your comment, much appreciated. My assumption is that the Q3 treats ISO "like metadata" as you pointed out. The video from the Leica Store in Nuremberg which I linked in the infobox also suggests that. So your comment is spot-on. Your point on "scaled down" images to be considered too is also highly relevant, I will include this next round in my test series. The smaller resolution already contributes by itself to a better signal-to-noise ratio for sure. On highlight recovery, I need to look into it to see if there is any difference with clipping in low vs high ISO images. In any case thanks for the great input and food for thought!
Thank you for the video… I have Q2 Monochrom, I shoot events in bars / nightclubs low light high ISO a lot (can’t use a flash) usually not going past ISO 25,000 but sometimes I need 50,000 ISO .. how do you think Q3 would compare at those high iso to Q2M ?
Amazing stuff and some interesting conclusions. Thank you for posting this. It does leave me with a question though. DR is very important. Good to see that there is no difference in shooting with high iso or pushing in lightroom. Very important to conclude the M dng setting offers best DR,. I gues one shoul you use L dng when you want to crop....I would like to know how this sensor containes highlights. The thing with the Q2 is, you have to underexpose a bit and push the shadows. In pratice how does the Q3 compare? Onze again thanks a lot! Waiting for my Q3......
Many thanks, Steven. The Q3 can easily underexpose by 5 stops and recover all information and color very well out of shadows. Overexposing works in my experience up to 3 full stops, thereafter highlights will be clipped and you lose information. But I typically use rather underexposure than overexposure although signal-to-noise is always best at the correct exposure. What comes in nice and handy is the highlight weighted metering setting. Then you always have the highlights perfect and then can recover information out of too shadowy parts of the image later in post. The Q3 is a fine and highly capable camera, I love it already after only a few weeks :) Crossing fingers you get yours soon!
Matt, I just watched the Video from the Leica Store in Nürnberg, very very good and extreme well explained, thanks for that link. It's more or less what I prayed over the years since I was confronted with this subject in my Fuji days. I also have to admit that You (your channel) are the only person I can discuss this stuff on this level and I am happy about that. What we also can learn from this excellent video, straight from the beginning, is that photography with a digital camera is still ANALOG PHOTOGRAPHY. More light equals more output voltage and besides that further down the (electrical) road there is a A/D convertor to prove this point again. In the video they show an example of a dark area that had more chroma noise in the 200 ISO version ( with added exposure in LR) than the 3200 ISO version (50:50) Now this is exactly what I wanted to point out when you use LR to enhance exposure with the exposure slider and C1 with the lightness slider. In the video at 32:15 the left guy makes a very important comment about the sensitivity of different coloured LEDs. The separation of colours in a sensor is done by the colour filter, RGB, but the fact is that R , G and B have different measures of attenuation of the light. So further down the (electric) road this has to be compensated (levelled) by increasing the the R, G an B channel separately to make it even again. Now there are companies that make you BELIEVE that removing the colour filter will make your camera a monochrome version of itself. This is not completely true, yes the an output is monochrome only by removing the filter , the attenuation per diode is also removed, HOWEVER later on the road the attenuation will still be applied so your output is not levelled. This can only be done by altering the firmware or any other internal setup to let the light per diode pass through 1 to 1 Then the right guy made another remark in the video I would like seriously debunk but that will "open up a can of worms" would rather save it for another day. Okay Matt rest me only to give you the link to the Leica Camera USA channel where Leica lens developers talk about quality of Leica lenses and in particular the SL (APO) lenses It's 2 years old but still very relevant: Search for - Leica Tech Talk: For the Love of Lenses Hope you enjoy. Greetings from a sunny Amsterdam, although I have to do some work today, but I don't mind, it's for myself, not for somebody else ;-)
Thanks RS, the Nuremberg Leica video is indeed very well done and very sympathic folks there :) Its a shame that their video had until yesterday only around 1k views, I hope my link will give them some traffic. I will look into the video you referenced, thanks for pointing out. Have a nice day - despite work for yourself - and greetings back to Amsterdam from very sunny Zurich ...
Amazing. I skipped some of the analysis, but followed most of the comparisons. Quick question. Why didn’t you try or show a higher level of noise AI in LR? Would there have been even more distortion? Please explain. Thanks! Geoff
By experience in LRC, 50% noise reduction is often the right balance between a cleaner image and still keeping good details. That's why I went for 40% and then for 50% noise reduction in Lightroom in the video. I worked with LRC on many images since they implemented the new noise magic button and 50% is for me the maximum. Beyond that images become a bit smeary and lose too many details.
Thank you. Very much looking forward to your similar analysis of the m11 (and maybe someday m11m)! Will be extremely interesting if there is a secondary independent threshold among the ISO settings (i.e. 3200 or 6400) where there is a second step of invariability OR where the invariability ends due to gain if there is only one linear invariability threshold for the m11 sensor. Great work!
You made difficult math into something I could follow. I hope there's no final exam. On a practical basis, would you maximize detail with DNG-L unless you needed the extra sensitivity? How often do fifteen stops vs. fourteen stops show up in prints, let alone on screen?
Thanks, great if the math in the video was not too cryptic :) The difference in practical terms between 14 and 15 stops of dynamic range is not very high and will make a small difference only in scenes with very difficult lighting conditions. And for prints you will anyway optimise the image by software, I have a whole suite of tools for post-processing and will use a longer workflow for demanding prints in terms of print size.
Just a quick note, going from 14 stops to 15 stops is 2x (i.e. 100%) more dynamic range, not 7% more. This is because the scale is logarithmic to base 2.
Fantastic! Thank you for an educational video. I have one question that I hope you have time to answer: how did you determine the upper range of the “native” ISO for the Q3 (or for any camera sensor)?
Very good lesson. I am awaiting my Q3 which is on order and I look forward to using it. Thank you for a very detailed video for us. I just subscribed and will follow you. Thank you again.
Matt, EXCELLENT VIDEO, no lie detectors went of here so everything you said is solid an true ;-) I watched the whole video and found it very interesting. As you know I advocate shooting low ISO and take advantage of the ISO Invariant Sensor, when available Love the calculations and your Python script (use Python a lot for almost everything), very very interesting HOWEVER I have some commands but more in the additional way (things that I missed) so you can breathe in peace and I hope you read my info: 1. What does ISO in Digital Cameras stand for: Where as ISO in film stands for the sensitivity of a film in a digital camera it does not represent the sensitivity of the sensor. A shot taken with higher ISO than base will be shot in base ISO and after the shot is taken a gain will be implied to boost the exposure. This is done by a tailor made algorithm for the camera sensor combination. 2. ISO Invariance, as you explained very well this is the opportunity of adding exposure in post with the same results as in camera iso boost, however in post we have more control cause we can adjust by shot, by looks and in camera it will be on average (still very good). I practice ISO invariance editing from my from Fuji days 10 years ago, with success cause the Fuji sensors excelled in that. Nowadays almost all brands have more or less ISO invariant sensors except Canon (they work with a different algorithm and there for it is better to rely on their higher ISO levels In those days I used Iridient developer, a far out better RAW editor for Fuji. 3. Capture one over Lightroom, as my Mantra goes, use C1 for those things cause the Ligthness slider has a far better algorithm to boost Exposure. 4. Now in your examples it was still reasonable light, but I used in pitch dark situations where I shot sceneries with a small amount of little streetlights and then C1 will give even better results. In examples where you showed Chroma Noise in the dark/black areas, C1 lightness will give you so much better control. 5. I already knew that the 36 MP position would give you the best Dynamic Range and I have seen it in pictures with half shadow and half sun reflected on white buildings. 6. For noise reduction and sharpness (which I don't use and is always set to zero in all my RAW editors), I rely on third party Applications dedicated to solve those problems, all available by a plugin in PS. I agree on your final conclusion that the Q3 is one of the better cameras out there. As you know 24MP is my favourite resolution but since I have seen the resolution the 36MP sensor setting, I am in doubt to chance my mint Q-P with all its boxes and what not and trade it for a Q3 This week I am busy but in the week after that I will go to my camera store and ask the for an offer. Who knows, but I also state: An OG Q is a hell of a camera and there are no reason for starters to buy in the Q system starting with an OG Q Matt, thanks for the effort to put this all together and I will link people in the future to show them the pure facts. And thanks for sharing, good weekend my friend !
Wow RS, I finally managed to get a true "like" from you :) Many thanks, my friend. There is so much good stuff in your comment that it is hard to pick on what I want to reply. But the ISO thought you shared is clearly an important one. I linked in the infobox below the video a live stream from the Leica Store in Nuremberg. In this video they explain nicely how the A/D converter in cameras found a different place in the value chain in cameras which have ISO invariance. Its in German language, so not suitable for everyone, but its a sympathic and well done video from Leica Germany/Nuremberg. Based on what I saw in that video and what I experimented myself, your statement on what ISO really is all about seems spot-on. By the way, surely I am aware that 24 MP is your favourite resolution :) But the Q3 with M-DNG with 36MP could be a candidate you might want to consider :) Have a nice weekend!
@@mathphotographer hehe I always thumb you up. I speak German since I am 4 and always took German as language choice in school. No learning and always best of class. Worked for a lot of German companies in NL. I'll check it out. There is a great American or Canadian video where 2 of 3 Leica lens designers claim the SL lenses are the best lenses they make(d). I'll look it up for you and share the link . Great interviews . Great taliking to you
Good analysis. But for pixel binning, how does in-camera binning compare with binning in Lightroom/Photoshop? Is the in-camera function just a convenience? And how does the DNR improvement compare with RAW files that start with more than 8 bits? More effect, or less? It’s actually remarkable that they can do pixelbinning with RAW files, which have the Bayer filter, especially since the binning is not by an integer number of pixels. Just thinking about it makes my head hurt. Do they demosaic the image, then scale it down, then remosaic? That would be crazy. Note that digital zoom affects only the JPEG and leaves the DNG whole, but DNG resolution (“binning”) does affect the DNG. There may be a hint in that they talk about the sensor being multi-resolution, this indicates there is some hardware there, it’s not just math in the processor. It’s impressive.
Absolutely agree, what Leica does here is a masterpiece and likely not only software but also some hardware tweaks they do not disclose to the public. I do not have all answers yet, still exploring and learning by experimenting and observing. And this video here was based on some work I did for myself to better understand the L/M/S-resolutions. And then I thought why not sharing with everyone on my channel, together we might find out more interesting stuff and then share :)
@@andersvinberg144 In the new LFI magazine Leica now says that the sensor in the Q3 is essentially the sensor from the M11. I will run these tests also for the M11 and M11 Mono in the next weeks.
Amazing content again :) Actually I study all my Hasselblad x2d functions with your videos appreciate . I have some questions , had Hasselblad x2d unbelievable quality images , can I get same image quality forhasselblad black and white with 100mp vs Leica b&w ? another question wanna get Leica for street photography cuz I like vintage looks , black and white and size vise so for Leica which one you recommend for images quality q3 or m11 , m11 mono ,sametime never used manual focus before but tried don't think that much difficult
I will do a video about the new Hasselblad X2D firmware soon, the X2D just got lots of new features. Now to your question: the Q3 is one of the best always-with-you cameras you can buy today. So if you are ok with having a fixed lens with 28mm, the Q3 is fully recommended to become your all-day always-carry camera :)
Danke für den tollen Erklärungen. Eine sache die in deiner Berechnung noch ingeressant gewesen wäre. was passiert mit den verschiedenen werten wenn du das 60Mpix Foto in ein 36,5 umgewandelt hättest. Hier müßte das Rauschen im bild durch die zusammenfassung der pixel eigentlich besser werden und die Frage ist in wie weit sich dann die werte annähern. Interessant ob sich dadurch zu sehen wäre auch ob sich die dynamik der bilder dann auch erhöht. Vielleicht magst du das ja noch einmal in text hinterher schieben. LG
@@mathphotographer You're welcome! What are your thoughts on the Q3 EVF in terms of brightness? I shot with the Q3 at Photoville in New York a few days ago, and I am seriously thinking of getting one. However, compared to the OVF in my Nikon D810, I felt like the Q3's EVF was a little dim in bright sunlight. Best, Stefan.
Outstanding presentation owing to your rigorous treatment of what is generally handled only subjectively. It highlights the progression of image quality on many fronts - both at Leica in terms of invariance in sensor readout vs ISO setting, but also in the new Lightroom AI based noise reduction - which I too have found to be game changing. I agree with others and eagerly await the next SL installment being a current SL2-S shooter. One question not answered - do you see the same ISO invariance and reduction in noise reported in this study on the M11, both standard and monochrome?
Excellent question, John. From side-by-side comparison of images I did not find pixel binning being effective on the Leica M11 so far and I also mentioned it in previous videos on the Leica M11 but what I presented here in this video is some tech I just recently explored and surely will also apply to the M11 to see what the results are :)
@@mathphotographer As you suggest, I'm eager to learn if applying the same rigorous approach that you used with the new Q3 will show the same benefits with the M11 for ISO invariance as well as pixel binning. Based on my experience with wide dynamic range conversion of analog signals in scientific instrumentation, achieving ISO setting invariance carries the implication that sensor readout noise is now very low, as it would otherwise add a bias that would scale disproportionately across ISO settings. I could imagine that this could also be making an indirect contribution to the pixel binning process as well. Is it possible that the introduction of the new Maestro IV processor is part of the equation?
I've now jumped out to the link you've provided "Online Talk im Leica Store Nürnberg - ISO Invariante Sensoren". Luckily German is not a problem for me. I'll watch this next - and I'm sure it will give me great info in this direction.
If the Q3 has same sensor as the M11, why aren't the results the same? In other words, why isn't there an increase in dynamic range in medium files in the M11?
Obviously...My point is that the M11 and the Q3 are very similar, so what exactly is the reason you are not getting the increased dynamic range with the medium file size? The sensor is the same as is the method of reducing the file size. Is it something that the M11 would be expected to have with updates?
@@pnw_md Thanks Steven. First of all, Dustin is right: its not only about the sensor, its also about the CPU (Maestro 4), the A/D converter and where its sitting in the image processing chain in-camera and then its also about the algos implemented. Second, I will apply the toolkit shown in this video to the M11 next weeks to see if pixel binning is effective in the M11 because by pure side-by-side image comparison I did not find true pixel binning in the M11 so far.