Which resolution do I use?

[maxfairclough]

Advertisement (gone after registration)

I was told by someone at Leica that bringing the camera down to 36MP increases the dynamic range. I am not sure how well known this is so I could see that benefit.

Edited August 29, 2022 by maxfairclough

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[SrMi]

3 hours ago, maxfairclough said:

I was told by someone at Leica that bringing the camera down to 36MP increases the dynamic range. I am not sure how well known this is so I could see that benefit.

That is true only when viewing at 100%, but not at the output level (print, screen). Leica has said that resizing in the post has the same effect.
Summary: 36MP does not increase DR in practice.

[FlashGordonPhotography]

10 hours ago, maxfairclough said:

I was told by someone at Leica that bringing the camera down to 36MP increases the dynamic range. I am not sure how well known this is so I could see that benefit.

What happens here is that by decreasing the resolution you are effectively making the pixels (virtually) larger. This reduces the noise floor which give the impression of more *usable* DR at a pixel level. But it doesn't matter if it's done in camera or in post. The improvement from 60MP to 18MP is just under a stop. It won't make or break a great photo.

Leica have not said whether it's hardware or software binning. Likely it's software. Mobile phones often use hardware binning to do the same thing. Reduce noise. Any phone with a 48MP sensor that outputs a 12MP file is doing exactly this.

Gordon

[maxfairclough]

On 8/30/2022 at 1:08 AM, SrMi said:

That is true only when viewing at 100%, but not at the output level (print, screen). Leica has said that resizing in the post has the same effect.
Summary: 36MP does not increase DR in practice.

 

On 8/30/2022 at 7:59 AM, FlashGordonPhotography said:

What happens here is that by decreasing the resolution you are effectively making the pixels (virtually) larger. This reduces the noise floor which give the impression of more *usable* DR at a pixel level. But it doesn't matter if it's done in camera or in post. The improvement from 60MP to 18MP is just under a stop. It won't make or break a great photo.

Leica have not said whether it's hardware or software binning. Likely it's software. Mobile phones often use hardware binning to do the same thing. Reduce noise. Any phone with a 48MP sensor that outputs a 12MP file is doing exactly this.

Gordon

Thank you both! That explains a lot.

[Jared]

I was curious whether the recent firmware updates had changed anything in terms of the effectiveness of the built-in "binning" in the M11 and whether there were any consistent differences in SNR depending on the mode used, so I decided to run a basic test. It isn't a particularly rigorous test, but it should be enough to find any obvious SNR benefits to the binning vs. simple down-sampling in post.  For reference, I am using firmware 1.5.0.1 on my camera which is, I believe, the most current. Here is what I did:

• Set ISO to 6400 on my M11 (since benefits to hardware binning are generally in the area of read noise, I wanted to choose a gain that was high enough to ensure read noise wouldn't be swamped by shot noise, as it might be near base ISO where exposure durations would be much higher)
• Set the lens wide open
• Chose a fixed shutter speed of 1/500s which was a "correct" exposure for the scene--the both shadows and highlights were right on the edge of clipping
• Shot a large DNG, a medium DNG, and a short DNG of the identical scene
• Brought the three images into Lightroom where I zeroed out all sharpening and noise reduction (though who knows for certain what Lightroom is doing even with these values zeroed out)
• Chose the "Done Monochrome" profile for all three images
• Exported the three images to Photoshop

I wanted to compare the 61mp image down-sampled to 36mp to the 36mp image, and I wanted to compare the 61mp and 36mp images down-sampled to the 18mp image. I used three different down-sampling algorithms and measured SNR for each. The three methods were bicubic, bilinear, and nearest neighbor. I measured SNR in decibels for a 100x100 patch in each of the images using 20*log((mean signal)/(standard deviation)) for calculating SNR. This won't be exact since the standard deviation will actually differ slightly from the noise value, but this should be pretty close, at least good enough to identify any trends. Here are the results:

61mp down-sampled to 36mp, bicubic, vs. 36mp "medium" file

1. Highlight, large file: 37.0dB
2. Highlight, medium file: 38.7dB
3. Mid-Tone, large file: 24.7dB
4. Mid-Tone, medium file: 26.3dB
5. Shadows, large file: 8.82dB
6. Shadows, medium file: 10.3dB

61mp down-sampled to 36mp, nearest neighbor, vs 36mp "medium" file

1. Highlight, large file: 36.3dB
2. Highlight, medium file, 38.7dB
3. Mid-Tone, large file, 23.9dB
4. Mid-Tone, medium file, 26.3dB
5. Shadows, large file, 8.12dB
6. Shadows, medium file, 10.3dB

61mp down-sampled to 36mp, bilinear, vs 36mp "medium" file

1. Highlight, large file: 37.8dB
2. Highlight, medium file: 38.7dB
3. Mid-Tone, large file: 25.5dB
4. Mid-Tone, medium file: 26.3dB
5. Shadows, large file, 9.77dB
6. Shadows, medium file: 10.3dB

61mp and 36mp down-sampled to 18mp, bicubic, vs. 18mp "small" file

1. Highlight, large file: 38.4dB
2. Highlight, medium file: 40.2dB
3. Highlight, small file: 38.9dB
4. Mid-Tone, large file: 25.8dB
5. Mid-Tone, medium file: 27.7dB
6. Mid-Tone, small file: 26.2dB
7. Shadows, large file, 10.4dB
8. Shadows, medium file: 11.6dB
9. Shadows, small file: 12.3dB

61mp and 36mp down-sampled to 18mp, nearest neighbor, vs 18mp "small" file

1. Highlight, large file: 36.0dB
2. Highlight, medium file: 38.8dB
3. Highlight, small file: 38.9dB
4. Mid-Tone, large file: 23.8dB
5. Mid-Tone, medium file: 26.4dB
6. Mid-Tone, small file: 26.2dB
7. Shadows, large file, 8.36dB
8. Shadows, medium file: 10.7dB
9. Shadows, small file: 12.3dB

61mp and 36mp down-sampled to 18mp, bilinear, vs 18mp "small" file

1. Highlight, large file: 37.7dB
2. Highlight, medium file: 40.5dB
3. Highlight, small file: 38.9dB
4. Mid-Tone, large file: 25.3dB
5. Mid-Tone, medium file: 28.0dB
6. Mid-Tone, small file: 26.2dB
7. Shadows, large file, 9.90dB
8. Shadows, medium file: 11.9dB
9. Shadows, small file: 12.3dB

So, what do I notice in terms of trends? Well, at least the results are pretty consistent. First, I'll note that to the eye on a good 4K monitor the differences in SNR at any given resolution were not easily visible, whether that resolution was achieved by shooting a "binned" file or by down-sampling in software. I suspect these differences are large enough that if I were pushing the files harder they would have been visible, i.e., if I were pulling up shadows a bunch or making major exposure corrections. But for an image that has minimal processing? You'd never notice the difference for any given resolution of image. White points and black points were not quite identical, but they were very, very close, so any difference in overall dynamic range would come down to shadow noise and recoverable details there. 

If I look at the actual measurements, there is a consistent, if slight, benefit to letting the camera bin the file to the required resolution rather than down-sampling in post, at least using the methods I described. This difference was pretty small, though, and there is one glaring contradiction. In highlights and mid tones, the SNR was better in 36mp "medium" images that had been down-sampled to 18mp than in the native 18mp "small" files. I replicated this with a second set of three images, by the way, and got the same result. In the darkest shadows, though, the trend reversed and the 18mp "small" file was better than the 36mp down-sampled file.

End result? If I know I was going to be pulling detail out of extreme shadows and that I was going to be limited in image quality more by noise than by resolution, I might actually choose to shoot at 36mp or even 18mp rather than at 61. Likewise, and as other have pointed out, if I know I'm just shooting snapshots that will be shared with friends and family I would probably shoot at 18mp just to save on storage and for the larger buffer. Also, for astrophotography where shadows noise is absolutely critical and sky rotation during the exposure will necessarily limit resolution, I would probably choose the 18mp or 36mp file size. However, for everything else I'd rather preserve the ability to crop my images while keeping as much detail as I can, so I'll use 61mp "large" files as my default, and will down-sample as appropriate (and get most--but not quite all--the benefits of in-camera binning.