GMCPhotographics said:

If the corners are software re-generated using AI

Click to expand...

blackcoffee17 said:

Im fine with digital corrections as long as large portions of the image are not replaced by AI generated data.

Click to expand...

I have no idea where these suggestions that the digitally corrected corners are filled with AI-generated image data come from, but I wish they would stop. That’s not how it’s done. No manufacturers that design lenses with an image circle smaller than the sensor do that, or have ever done that.

It’s the sort of thing that some influencer pushing an agenda would make up and promote as factual. And some people believe it, and repeat it. The whole idea is so silly that it gives me a headache. I’d take some Tylenol, except I heard that causes autism.


Edit: I found at least one place such drivel is coming from or at least being parroted locally…

GMCPhotographics said:

DXO's AI corner filling isn't adding true detail it's generating / guessing what might have been there. AI is capitalising on the fact that the lens is designed a bit wider angle of view so there is room to software correct and still make the stated focal length.

Click to expand...

EthenJ said:

It sure doesn't feel like RF is less expensive than EF glass

Click to expand...

Direct comparison with the EF era is tricky because the overall size of the market has shrunk so much. The economics of the thing must have changed a fair bit over the past couple of decades.

Lovely flower.

EthenJ said:

Or look at the EF 100-400L vs the RF 100-400. The RF is much lighter, yes, but the EF is sharper and cheaper. And that's considered one of the best value RF lenses around.

Click to expand...

The current EF 100-400mm ii L costs new here £2599 and the RF £749. The original EF 100-400mm was discontinued in 2014 and was well into a 4-figure price. My RF 100-400mm is not as sharp as my EF 100-400mm iis were but is sharper than my original EF.

Rafał said:

You missed the point that optical correction of distortion is not free - it also introduces aberations and losses in quality.

Click to expand...

I make this point frequently in these discussions, but I guess some people’s views on the matter are too distorted to accept that fact.

Rafał said:

There was an interview with Sigma engineers who said that in many cases correcting distortion and vignetting digitally results in better sharpness than doing it optically.

Click to expand...

Would be an interesting read, if you have a link to it.

Traveler said:

24 MP is the exact APS-C crop resolution of a 54 MP full-frame sensor. However, Canon uses a 1.6x crop factor. The 54 -> 24 MP reduction corresponds to a 1.5x crop factor.
So if it was not Canon, I'd be pretty sure that someone saw a cropped version of the image.

I don't think dual resolution makes much sense. It would have to be pixel-binned (something like 80/20 MP, four neighboring pixels would share the same color in the Bayer mask). In that case, I don't think it would bring much of an increase in detail in the 80 MP image.
If the binned pixels don't share the same microlens, then you'd still have to store data from all the pixels to recreate the color information.

Click to expand...

If this isn't totally made up, they could be using something more complex than the quad-bayer array like smartphones normally use for pixel binning.

The simplest scheme that fits the rumor would be some kind of 3x3 pixel array that could be sampled as either 9 pixels or 4 pixels, lightly software corrected for the full 9-pixel resolution. But I'm not qualified to even guess what they would look like.

If that's the case, all our existing editing software will need updates to process RAWs. We might be able to get hints by peeking at changelogs.

Lens corrections are great! It's a transparent tool that just makes lenses lighter and less distorted, but:

So, in Canon’s defense, this allows them to produce lenses that are smaller, lighter, and at times, less expensive than if they didn’t take that compromise.

Click to expand...

"Less expensive" is the key.

It sure doesn't feel like RF is less expensive than EF glass, expecially in the budget range. For example, Canon sells the 24-50 kit lens for $350. I got it for $89 used... and even at that price I'm feeling some buyers remorse :/

Or look at the EF 100-400L vs the RF 100-400. The RF is much lighter, yes, but the EF is sharper and cheaper. And that's considered one of the best value RF lenses around.

Or, heaven forbid, look at uncorrected third party RF, like the Yongnuo 23/35/56 1.4. $220 each, and very sharp. Or the 85mm 1.8 for $320, or the Venus Optics ultrawides, or the Argus 0.95s, all of which have to get by without in-body corrections because Canon won't open their mount like almost almost every other manufacturer.

Note I'm not blaming the engineers here; they can't do anything about Canon's margin requirements.

If people don't want digital corrections they should just use film cameras ( ͡° ͜ʖ ͡°)

AlanF said:

Centre or edges?

Click to expand...

Edges! By the way, even TDP's Brian noticed some 14-35 sample variations which he kept mentioned for a short lapse of time on the lens' page.
Especially when using my favourite focal length, 35mm, I pay attention to obtaining the sharpest corners possible. When travelling. I don't use primes, but a 35mm TSE would be for me an instant buy.

Malm said:

I think that there is a larger niche for a manual fast and good 14mm RF lens as the 'astro' community isn't that small any more.
I theory Sigma can just add an aperture ring to their lenses and replace the AF by a MF system and sell them with an RF bayonet. Well, there might be reasons why they don't do it ...

Click to expand...

I believe that a Chinese brand is more likely to bring out such a lens as they don’t currently seem concerned about keeping canon happy for potential authorised lenses in the future….

But you are correct that sigma/tamron could port their existing lenses to EF or RF (EF protocols) but they choose not to. I would like to think that any r&d cost would be relatively low vs potential sales but I don’t have the data to support that viewpoint

Del Paso said:

Rejecting software correction implies rejecting also DXO, LR etc… editing.
What matters is how good the picture you obtain is, whether optically or electronically corrected. Period!

Click to expand...

It has to be a limit though. Some vignetting and distortion correction is acceptable but i hope we won't see 20% of the sensor area replaced by AI generated image data because the final image is still "good".

Im fine with digital corrections as long as large portions of the image are not replaced by AI generated data.

Rafał said:

You missed the point that optical correction of distortion is not free - it also introduces aberations and losses in quality. There was an interview with Sigma engineers who said that in many cases correcting distortion and vignetting digitally results in better sharpness than doing it optically.

Click to expand...

Fixing corner vignetting due to optical shading is not easily "fixed" with a profile. I fthe profile is AI powered then it can clean the iso noise and re-generate the missing details. But a simple mathmatical corner stretch and then applying 4.5 stop white circle overlay over the far corners is only going to add 4.5 stops of noise in the corners to your raw file. Shoot your Raw landscape at say 400 iso, your corners are already profiled at 6400+ iso before you even start your post prod on the image. There's no magic bullet with camera generated lens profiles.

The Digital corner profiling discussion is as complex as the use multitude of case scenarios from the users who shoot with these lenses.
If the corners are software re-generated using AI then there is no loss of detail or Dynamic range. You should be able to push the exposures and oggle the fine detail in the corners and be happy with the results. HOWEVER.....Canon doesn't employ AI software based corner regeneration, what they do is stretch the corners and use the raw file's DR to bump the corner exposure to that of a perfect lens. This is fine if you don't mind loosing a little bit of resolution, maybe even a bit of focal length and a lot of corner DR.
Most Canon UWA lenses already give us a slightly wider angle of view to their acclaimed wide focal length marking. So a RF 14-35 F4L lens is closer to 13.5mm when uncorrected, the focal value of the lens corresponds to the corrected focal length. Another example of this is the sublime RF10-20mm f4L, it's notably wider when uncorrected. So if you take the uncorrected RAW image and run it though software that can AI regenerate the corners, you get an even wider angle of view than the stock lens with Canon correction applied.

Where the non AI software corner correction falls down is in low light landscapes where the rest of the scene is within the reasonable DR range of the sensor. The corners may have been pulled by the profile before you even start to post process the image. If you shoot on a tripod at 100 iso for a super clean image, the far corners might already be at 1600iso noise levels becuase of the correction profile. Once you pull the shadow details a bit (say two stops) the majority of the image will have super low noise at an equivelent of 400 iso...nice and clean. But these far corners will have an equivelent iso value of 6400 and will have a noticable amount of noise compared to the rest of the image. That's just a 2 stop pull. Ok you have AI noise correction, but this approach from Canon does introduce more work for lanscapers. Even a neat 100iso image will have grainy corners, you would need to stack multiple exposure blends to nullify the issue.
Or you crop the bat-sh1t crazy corners out...or use AI generation for the corner fills. It's the penalty for smaller and lighter lenses like the EF 10-20mm f4L.

mimbu said:

You can call it loyalty, or inertia, or whatever you wish. It's all the same thing. You keep buying from the same company even if there are potentially better options for some of what you do. A lot of people don't even educate themselves on what those options are! Like the author of the article this thread comes from, posting about how Sigma is dead to him because they don't exist on RF. Maybe expand horizons a bit.

Click to expand...

It's not the same though, is it?

As for education, I know it's impractical for me to change system (and I have seen nothing compelling that would entice me to do so) so why waste my time learning about the alternatives? It might be interesting for the sake of curiosity but that's about it.

Some people are always searching for something to make things better, some people meet the world where it is. Neither is necessarily superior (and most people do both at different points). Not everyone can or should be like you.

David - Sydney said:

If a good performance manual 14/1.4 or 14/1.8 on RF or EF mount was released then it would be a no-brainer for me. No different to my Samyang 14/2.8 decision in the past.

Click to expand...

I think that there is a larger niche for a manual fast and good 14mm RF lens as the 'astro' community isn't that small any more.
I theory Sigma can just add an aperture ring to their lenses and replace the AF by a MF system and sell them with an RF bayonet. Well, there might be reasons why they don't do it ...

Rafał said:

You missed the point that optical correction of distortion is not free - it also introduces aberations and losses in quality. There was an interview with Sigma engineers who said that in many cases correcting distortion and vignetting digitally results in better sharpness than doing it optically.

Click to expand...

I’d love to learn more about this; do you have links to those interviews?

Del Paso said:

While in "physical" my comparison, the 15-35 was noticeably better at 35mm, and at least as good at the other focal lengths...

Click to expand...

Centre or edges?

AlanF said:

Great analysis Richard! Go one stage further and compare the two lenses at 18mm, 24mm and 35mm at f/4, and you see from the optical limits charts that the 14-35mm has far, far less fall off on going from the centre to the extremes. So, for this pair of lenses, the optical correction at the shortest length causes far more damage at longer lengths than does the digital!

Click to expand...

While in my "physical" comparison, the 15-35 was noticeably better at 35mm, and at least as good at the other focal lengths...

Klendothir said:

I had both the 15-35 and the 14-35. I finally sold the 15-35 last week because it constantly stayed at home. The 14-35 is so much smaller, lighter AND a full mm wider where it really counts. I can definitely live with one stop less light at these lengths. In the worst case, I can crank up the ISOs (I use DxO PhotoLab anyway) And while I also have the 10-20, this one often feels too wide.. The 14-35 barely leaves my camera during a city trip.

Click to expand...

I agree, was my only lens on a 3 week hiking trip in the alpes last summer. The only thing i missed was a tele lens for the occassional wildlife. It is also good enough for auroras and okay for wide field astro, especially because of the convenience of having one lens. Now i am considering the 20mm for handheld astro/auroras but is not sure if its wide enough. Would love a third party manual in the 14-18 range with f2 or something

24 MP is the exact APS-C crop resolution of a 54 MP full-frame sensor. However, Canon uses a 1.6x crop factor. The 54 -> 24 MP reduction corresponds to a 1.5x crop factor.
So if it was not Canon, I'd be pretty sure that someone saw a cropped version of the image.

I don't think dual resolution makes much sense. It would have to be pixel-binned (something like 80/20 MP, four neighboring pixels would share the same color in the Bayer mask). In that case, I don't think it would bring much of an increase in detail in the 80 MP image.
If the binned pixels don't share the same microlens, then you'd still have to store data from all the pixels to recreate the color information.

Thank you @AlanF and @Click

AlanF said:

Great analysis Richard! Go one stage further and compare the two lenses at 18mm, 24mm and 35mm at f/4, and you see from the optical limits charts that the 14-35mm has far, far less fall off on going from the centre to the extremes. So, for this pair of lenses, the optical correction at the shortest length causes far more damage at longer lengths than does the digital!

Click to expand...

That's true, and it's much harder to get to 14mm as well in terms of optical design. The 14-35 has a slightly more zoom range, and because it's software reliant, the designers can engineer a better-performing lens overall.

Richard CR said:

OpticalLimits has recently gotten its hands on the Canon RF 14-35mm f/4L IS USM, which was announced way back on June 29, 2021. That in itself probably isn't newsworthy. So instead of talking about the review itself, I focused on a specific part of the review that caught my eye. This part of the review […]

See full article...

Click to expand...

Great analysis Richard! Go one stage further and compare the two lenses at 18mm, 24mm and 35mm at f/4, and you see from the optical limits charts that the 14-35mm has far, far less fall off on going from the centre to the extremes. So, for this pair of lenses, the optical correction at the shortest length causes far more damage at longer lengths than does the digital!

Rafał said:

You missed the point that optical correction of distortion is not free - it also introduces aberations and losses in quality. There was an interview with Sigma engineers who said that in many cases correcting distortion and vignetting digitally results in better sharpness than doing it optically.

Click to expand...

That's a fair point. I implied it, that by not worrying about the image circle as much, it allows the designers to focus on other aspects of the design, but yes, I didn't go into as much detail as I should have.