Industry News: Sony Alpha a7s III images leak ahead of tomorrow’s announcement

A4 at 300 DPI is 8.3 MP, and that's only needed if you're looking at the image from reading distance, e.g. a photo printed in a magazine.

And we know most photographers have their photos printed at least A3 size in such magazines as national geographics.
True enough. In my own work doing architectural photography though, I can see the limits of 30MP in a 5Div and R. I should always qualify my snarks.
 
  • Like
Reactions: 1 user
Upvote 0
Jan 29, 2011
10,675
6,121
My point is simply that the noise at the pixel level does matter.
Well that is a complete turn around and not what you have been saying for days. My point was that noise at the pixel level is irrelevant when comparing different density sensors, yours was you need to inspect both at the pixel level.
 
  • Like
Reactions: 1 user
Upvote 0
It's not only about high-ISO noise, it's about the noise in general at any given ISO. If I buy a high-MP camera, I'm interested in the pixel-level noise to be as little as possible because it determines the usability of native size images. I don't need a 45Mp camera to always downscale the images to 20Mp. Why would I need 45Mp if I only post to Instagram?

Also the pixel-level noise is important when I crop the images. When I crop but keep the print size the same (or target digital image size the same), I effectively magnify the noise, so again I'm interested in it to be as little as possible.

Honestly I'm not sure what we're even discussing here. You were talking about 'pixel level measurements', and I said we deal in images, not pixels. Nothing you've said changes that. I mean, you do you, but I do think you've demonstrated in the past a... shall we say, particular approach to noise and sensor measurement (in terms of choosing between cameras). I don't understand your position enough to know if I think you're incorrect, or if it's just differences of opinion. Of course most of us want the cleanest images we can get. But noise and DR can only meaningfully impact our work on an image level.

In answer to your question, why buy a 45MP camera only to downscale - well, not everyone wants to run multiple cameras at once. If I buy a high resolution body for extreme cropping of certain subjects, I wouldn't necessarily also have the budget (or space in my bag) for a lower resolution one for other purposes. And anyway, almost every use will involve downscaling images because nobody has anything approaching a 45MP viewing device, and very few people are regularly printing large enough and viewing from appropriate distances to fully utilise that resolution. Downscaling 45MP>20MP or whatever is going to create cleaner, sharper images than a native 20MP image, all other things being equal, because a) you can do NR on the full resolution image before downscaling, and b) the higher resolution sensor is capturing more detail. Though in practice how much difference this makes depends on a lot of factors beyond sensor resolution.
 
Last edited:
  • Like
Reactions: 1 user
Upvote 0
Hahaha. I’ve work with the ARRI Alexa before for a Netflix movie. I was being sarcastic.
 
Upvote 0
My point was that noise at the pixel level is irrelevant when comparing different density sensors, yours was you need to inspect both at the pixel level.

Pixel-level noise matters when you consider one or another sensor for your purposes.
It's also easy to compare sensors of similar megapixel count, e.g. R5, A7RIII, 5DsR, Z7.

When you compare sensors with different Mp count, such as R5 and R6, you take only one metric (e.g. dynamic range or noise) and deliberately downsample the R5's images thus losing 55% of information. 55%! As a result you're not comparing the sensors, you're comparing only the noise in normalised images. This comparison ignores the resolution and is only valid for cases where you downsample 45Mp to 20Mp. If you want to use all 45Mp from the R5, or downsample it to 30Mp, this comparison becomes totally useless.
 
Upvote 0
Honestly I'm not sure what we're even discussing here. You were talking about 'pixel level measurements', and I said we deal in images, not pixels.
Analysing pixel-level noise means viewing images without downsampling, like this. It wasn't about analysing separate pixels! :)
Why it matters for dealing with images - I tried to explain in the previous messages. I actually mostly agree with what you've described as your approach.
 
  • Like
Reactions: 1 user
Upvote 0
Jan 29, 2011
10,675
6,121
Pixel-level noise matters when you consider one or another sensor for your purposes.
It's also easy to compare sensors of similar megapixel count, e.g. R5, A7RIII, 5DsR, Z7.

When you compare sensors with different Mp count, such as R5 and R6, you take only one metric (e.g. dynamic range or noise) and deliberately downsample the R5's images thus losing 55% of information. 55%! As a result you're not comparing the sensors, you're comparing only the noise in normalised images. This comparison ignores the resolution and is only valid for cases where you downsample 45Mp to 20Mp. If you want to use all 45Mp from the R5, or downsample it to 30Mp, this comparison becomes totally useless.
No! You normalize, it doesn’t matter if you normalize up or down, the noise in normalized images is a direct comparison. I don’t understand your disconnect here, you don’t throw anything away and you don’t make stuff up, that isn’t what normalizing is doing, it is putting things on an even footing.

But you don’t need to resample up or down, just look at stuff at the same size not the same ratio (percentage). Full screen, thumbnail, print size it doesn’t matter, if the lower pixel image goes above 100% to view the same output size as the larger it is irrelevant, it is still a normalized comparison.

Online tools have to resample the higher resolution to a lower one so the display works, to make it simple they resample everything, high and low resolution sensors, to a level no camera will realistically go below. The 20mp camera and the 45mp camera are both downsampled to a common 8mp, but in your own home you don’t need to do that, just look at both on screen so any subject or detail in both files is displayed on your screen the same size. That is normalizing, no resample, no throwing away anything no making stuff up.
 
  • Like
Reactions: 1 user
Upvote 0
Jan 29, 2011
10,675
6,121
I find these two features remarkable:
- 15 stops DR
- 9.44 EV which is almost a 1 to 1 of the 12mp sensor
I would also assume that ISO performance (SNR) is exceptional. Sony is probably still ahead in sensor technology.
Apart from the fact that sample video shows no real world DR advantage over video from the R5 and it is not a 9.44 mp viewfinder it is a 9.44 million dot viewfinder which means it is a 3.2mp viewfinder.

Oh and it appears to overheat!
 
Upvote 0
No! You normalize, it doesn’t matter if you normalize up or down, the noise in normalized images is a direct comparison. I don’t understand your disconnect here, you don’t throw anything away and you don’t make stuff up, that isn’t what normalizing is doing, it is putting things on an even footing.

For such comparisons you don't normalise up if you want to get any reliable results. When you scale an image up, you add information that wasn't there through various types of interpolation. Now if you try to measure the noise, you'll measure not only the original noise but also your artificial interpolation data.

The 20mp camera and the 45mp camera are both downsampled to a common 8mp, but in your own home you don’t need to do that, just look at both on screen so any subject or detail in both files is displayed on your screen the same size. That is normalizing, no resample, no throwing away anything no making stuff up.

But what you described is pure resampling - only you don't convert the image files, but resampling happens before the images are displayed on the screen.
 
Upvote 0

bbasiaga

Canon Shooter
Nov 15, 2011
721
971
USA
A4 at 300 DPI is 8.3 MP, and that's only needed if you're looking at the image from reading distance, e.g. a photo printed in a magazine.

And we know most photographers have their photos printed at least A3 size in such magazines as national geographics.
I think its pretty clear from the R6 threads that most photographers have to crop a finch out of a 100m field of view and still have enough pixels to print it wall sized for their local museum/gallery. Minimum usable megapixels are about 10,000,000. ;)

All in jest folks...all in jest.
 
Upvote 0
Jan 29, 2011
10,675
6,121
For such comparisons you don't normalise up if you want to get any reliable results. When you scale an image up, you add information that wasn't there through various types of interpolation. Now if you try to measure the noise, you'll measure not only the original noise but also your artificial interpolation data.



But what you described is pure resampling - only you don't convert the image files, but resampling happens before the images are displayed on the screen.
You are still missing the point by staring at a tree when people want you to describe a forest. As photographers rather than scientists (I know there are some scientists here but give me some leeway here) we are interested in presenting pictures, be that a print on a wall, an instagram post whatever. All I am saying is the ultimate way of comparing two different pixel density sensors is to look at both at your intended output size. That is normalization and that is what will give you all the information you need two make a decision about IQ.

Knowing the amount of photons an individual pixel can catch is an interesting academic exercise but it tells you less than zero about the IQ differences you are going to see between the R5 and R6 (for example) for any given output.

The point is if I want to make a 24"x36" print should I get an R5 or R6? Everybody is going to say an R5 because you need the extra pixels. Ok but what if I have to shoot at iso 3,200 at f11? Then there is no difference between the two and I can prove that by normalizing both images to that size on screen, that is the only view that has meaning in the context of my personal need as a photographer making pictures. If I only want to print an 8"x10", I just view both on screen at 8"x10", if I want to make an instagram post I view both at max 1080px wide.
 
Upvote 0
You are still missing the point by staring at a tree when people want you to describe a forest. As photographers rather than scientists (I know there are some scientists here but give me some leeway here) we are interested in presenting pictures, be that a print on a wall, an instagram post whatever. All I am saying is the ultimate way of comparing two different pixel density sensors is to look at both at your intended output size. That is normalization and that is what will give you all the information you need two make a decision about IQ.

It's one of the ways to compare sensors, that's right. But it has limitation as in my message above. You're losing information from the higher-res sensor in this comparison, effectively providing a handicap. My goal may be print larger than the lower-res sensor can handle. Or have a room for heavy cropping.

The point is if I want to make a 24"x36" print should I get an R5 or R6?

At what ppi? Make it 200ppi and the R6 won't handle it.

Ok but what if I have to shoot at iso 3,200 at f11? Then there is no difference between the two and I can prove that by normalizing both images to that size on screen

Within certain limits you can print from both cameras the same quality prints, but you've chosen bad examples as the R6 just won't be able to handle such target print size at an acceptable resolution (e.g. 200ppi).

If I only want to print an 8"x10", I just view both on screen at 8"x10", if I want to make an instagram post I view both at max 1080px wide.

That's a valid comparison but you need to understand its limitations. It's like comparing Ferrari with max. speed of 340 kmph and Corolla with max. speed of 180 kmph. You put them both in the school zone and compare performance. Then you conclude there isn't much of a difference apart from different ergonomics...
 
Upvote 0