What’s next with full frame sensors?

[FlashGordonPhotography]

Advertisement (gone after registration)

4 hours ago, SrMi said:

Dynamic range can not be compared unless you use the same measuring technique. E.g., Sony claims to measure 15 stops for its 60MP sensor, while P2P measures less than 12 stops. There is also the video vs. still, which is entirely different, but that is the only thing I know about video :).

The max DR in still photography has not changed since 2013 (Sony a7R announcement). The only significant advancement was the addition of dual conversion gain, which helped a lot at higher ISOs.

Yes but *usable* DR has. This is why site that measure DR are pointless.

The SL3 sensor allows much more file pliability, especially in the shadows, than the SL2. Actually, slightly more than the A7R5 as well. There’s differences in the processors, micro lenses and exposure algorithms and this does make a difference. My Sl3’s have about a usable stop more DR than my A7R3. And if I take DXO into account my 2024 A7R3 files have more usable DR than my 2018 A7R3 files.

People like to quote DR figures as if they are the be all and end all. *Usable* DR is hat really matters. There’s always a stop or two of measured DR that is not worth the conversation. Measured DR may not have changed in 8 years but usable DR certainly has. Ata a pixel level the SL3 basically matches my S5 and S5II.

Gordon

p.s. someone mentioned bigger pixels blah blah blah. High resolution sensors do NOT have less DR than lower res sensors of the same size. This is a myth. Downsample to the same resolution and the differences evaporate. And I’ll put the SL3 and M11 against any sensor on the market for IQ and unstable DR, with the SL3 being slightly better and the M11 merely the equivalent of the best every other brand has to offer….

[Advertisement]

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

[FlashGordonPhotography]

16 hours ago, o2mpx said:

Currently, for travel landscapes, the X2D with 21/35-75/90/135+TC weigh a ton; the SL2 plus 16-35/24-90+Sigma 105+TC isn’t substantially better. 

Would prefer higher than 60mp FF sensor that can support cropping, and carry only 1 or 2 zooms. 

Agreed. This is why I have my X2D for primes and the GFX for zooms. I can build a reasonably lightweight kit with a GFX100S(II) and three zooms. Since there’s only one (spectacular though) zoom for the Hasselblad, it’s much harder. OTOH an X2D with a 55V weighs about the same as an SL3 and non APO Summicron, with a leaf shutter and better IBIS.

I would love to see a 20-35 and 80-200 from Hasselblad. Then I’d get a 907x/100C and carry a small tripod. Landscape heaven, for me. The Hasselblad has much better long exposure implementation and that would suit if they’d drop a couple of zooms.

I have to say though I prefer *using* the SL3 over the GFX, even though the latter has better IQ. I like that I can have a different kit weight by just changing lenses around. The GFX100II is a sensational image maker but an uninspiring camera, for me. I use it when I *have* to. The Sony A1 is more fun to use. Personally I find the SL3 and X2D the most satisfying cameras to use by a long way. I get it doesn’t matter to some but it does to me.

Gordon

[o2mpx]

For me, analysis on equipment within the context of the use case points to practical challenges. Typically I’m part of a group tour on a hike or visiting places of interest, without time to find the ideal vantage spot, nor wait for better light. For this use case, have found zooms rather than primes, higher resolution sensors to support cropping, ideally lower weight and less bulk, to be desirable. Unfortunately, these 3 criteria work against each other. 

The X2D+lenses have the sensor and IQ, terrible weight and bulk. The SL2 slightly less bad on weight and bulk, but half the resolution of the X2D. Perhaps the SL4 might do the trick if it gets a 75/80mp sensor, output DNG’s instead of needing to fuss with Phocus and 3fr’s and 3fs files. 

Edited July 7, 2024 by o2mpx

[evikne]

Some photographers have a philosophy of shooting all images at Daylight WB. That way, all the natural variations in light throughout the day will be more apparent. In much the same way, I imagine that with the sensors of the future, you can shoot all images at the same ISO value. The sensor will still capture everything from the darkest to the brightest in every image. We're almost there already with today's ISO-invariant sensors, and I expect this can be further improved. Perhaps the cameras of the future won't need any ISO controls at all?

[jaapv]

On 7/5/2024 at 4:38 PM, o2mpx said:

Do we expect to see higher than 61mp sensors in the next gen? How close will it get to 100mp used by MF Fujis and HBs?

When/if the next iteration is close to 75mp, would it justify passing on MF and sticking with FF reduced weight and bulk?

Maybe slightly more? The real resolution jump will be in five to ten years. 

[jaapv]

13 hours ago, FlashGordonPhotography said:

Yes but *usable* DR has. This is why site that measure DR are pointless.

The SL3 sensor allows much more file pliability, especially in the shadows, than the SL2. Actually, slightly more than the A7R5 as well. There’s differences in the processors, micro lenses and exposure algorithms and this does make a difference. My Sl3’s have about a usable stop more DR than my A7R3. And if I take DXO into account my 2024 A7R3 files have more usable DR than my 2018 A7R3 files.

People like to quote DR figures as if they are the be all and end all. *Usable* DR is hat really matters. There’s always a stop or two of measured DR that is not worth the conversation. Measured DR may not have changed in 8 years but usable DR certainly has. Ata a pixel level the SL3 basically matches my S5 and S5II.

Gordon

p.s. someone mentioned bigger pixels blah blah blah. High resolution sensors do NOT have less DR than lower res sensors of the same size. This is a myth. Downsample to the same resolution and the differences evaporate. And I’ll put the SL3 and M11 against any sensor on the market for IQ and unstable DR, with the SL3 being slightly better and the M11 merely the equivalent of the best every other brand has to offer….

I am not very optimistic about significant increases in noise performance AKA effective Dynamic Range. Most types of image noise are due to physics- the purity of the silicon used, the fill factor of the pixels ( increasing the capture surface of the sensor by having less circuitry obscuring the capturing surface of individual pixels) and reducing the sensitivity variation between pixels. Unless there is a revolution in light capturing technology the present boundaries are being pushed already. Black noise by background radiation will always be present, electronics will always generate background noise. 
As an analogy old sensors only managed to fill the bucket 50%, modern ones  maybe 90% but it is impossible to go beyond a completely filled bucket. 
 

https://en.m.wikipedia.org/wiki/Image_noise

The future of getting smoother images in low light can basically only be by the application of AI technology which is being developed right now, and moving it from postprocessing into firmware. The advantage of high resolution sensors might be the generation of more data for AI to work with but that will depend on the  (lack of) meaningfulness of those data. 
Another but more esoteric way might be multiple exposures stacked in a very short time but that runs into issues like sensor readout time vs movement of the subject analogous to high-resolution modes on newer cameras, or maybe multiple layer sensors but those are have their own problems (like how to focus on a many-layered  sensor, and acceptance angles) 

However, there are very smart people out there and I might be proven wrong. 
 

[FlashGordonPhotography]

Advertisement (gone after registration)

5 hours ago, jaapv said:

I am not very optimistic about significant increases in noise performance AKA effective Dynamic Range. Most types of image noise are due to physics- the purity of the silicon used, the fill factor of the pixels ( increasing the capture surface of the sensor by having less circuitry obscuring the capturing surface of individual pixels) and reducing the sensitivity variation between pixels. Unless there is a revolution in light capturing technology the present boundaries are being pushed already. Black noise by background radiation will always be present, electronics will always generate background noise. 
As an analogy old sensors only managed to fill the bucket 50%, modern ones  maybe 90% but it is impossible to go beyond a completely filled bucket. 
 

https://en.m.wikipedia.org/wiki/Image_noise

The future of getting smoother images in low light can basically only be by the application of AI technology which is being developed right now, and moving it from postprocessing into firmware. The advantage of high resolution sensors might be the generation of more data for AI to work with but that will depend on the  (lack of) meaningfulness of those data. 
Another but more esoteric way might be multiple exposures stacked in a very short time but that runs into issues like sensor readout time vs movement of the subject analogous to high-resolution modes on newer cameras, or maybe multiple layer sensors but those are have their own problems (like how to focus on a many-layered  sensor, and acceptance angles) 

However, there are very smart people out there and I might be proven wrong. 
 

I agree with you. I also think it’s inevitable that the NR applications will move from something like DXO to on camera. A dedicated NR processor, perhaps? Improvements in usable DR have happened from  my A7R3 to my A7R5/M11 and again with the SL3. These aren’t improvements in the silicon, as such but everything else in the pipeline. None the less those improvements are real and I’m enjoying them. I think, like Leica do, that we are closer to that limit now, also and it’ll be computing power that takes it to the next level.

The SL3 files are the most pliable small format files I’ve seen in a high resolution sensor. Basically like working with the S5II but with more data. It’s like working with a 60MP miniMF file. And that’s without DXO. I really don’t care how they do it. I’m just enjoying the ride.

Gordon

[SrMi]

37 minutes ago, FlashGordonPhotography said:

I also think it’s inevitable that the NR applications will move from something like DXO to on camera

I hope that does not happen (although some manufacturers already apply NR to raw data today). NR should be tunable in the post, and raw data should be as "virgin" as possible. The same is true for focus stacking (assembly) and panorama stitching.

I expect Adobe to implement AI NR, which does not require an extra file.

[jaapv]

I think that Lightroom Enhance is AI already. 

[SrMi]

2 hours ago, jaapv said:

I think that Lightroom Enhance is AI already. 

Yes, and many other functions. I used an unnecessary comma in #28, which may have caused a misunderstanding. Clarification:

I expect Adobe to implement AI NR without needing an extra file. That extra file is the main nuisance of using AI NR.

[FlashGordonPhotography]

13 hours ago, SrMi said:

I hope that does not happen (although some manufacturers already apply NR to raw data today). NR should be tunable in the post, and raw data should be as "virgin" as possible. The same is true for focus stacking (assembly) and panorama stitching.

I expect Adobe to implement AI NR, which does not require an extra file.

Raw files are getting further and further from actually being raw, with every camera release. I don’t expect that to change. NR is baked into the files already for most cameras. It’ll just get better with each iteration. You already have baked in lens profiles on the Sl bodies.

Gordon

[SrMi]

5 hours ago, FlashGordonPhotography said:

Raw files are getting further and further from actually being raw, with every camera release. I don’t expect that to change. NR is baked into the files already for most cameras. It’ll just get better with each iteration. You already have baked in lens profiles on the Sl bodies.

Gordon

Most cameras do not have NR baked in except in the highest ISOs. P2P runs NR detection algorithms and Pentax and Canon are exception, as they add NR at lower ISOs as well, though Canon seems to apply only slight NR.

To the best of my knowledge, lens profiles are not baked in on the SL bodies. They are added as metadata, which can be removed or turned off. The raw data is not modified. I believe that M cameras modify raw data in the corners depending on the detected lens.

Anyway, except for Pentax/Ricoh, the raw data is only very slightly modified.