Canon officially announces the RF 24mm f/1.8 Macro IS STM and RF 15-30mm f/4.5-6.3 IS STM

dirtyvu

EOS 90D
Jan 7, 2019
153
115
So do I, and although I've never used a Laowa Zero-D lens, it would be ironic to introduce geometric distortion to correct an optically barrel-free UWA lens, to correct the anamorphosis.
i went back to that review and it doesn't seem to be describing what you're saying. and then I went and looked up anamorphosis and it doesn't seem to have to do with the lens.

Here's what it says:

Based on DxO Labs' exclusive geometric correction technology, DxO ViewPoint allows photographers to easily and efficiently restore the natural proportions of the subjects in their images as an integral part of their workflow.

The deformation of subjects located on the edges of photos is due to the use of wide-angle lenses, and affects numerous situations, particularly social photography (e.g., events, marriages) and photo reportage. Thanks to its dedicated tool, DxO ViewPoint easily fixes this flaw, automatically restoring the natural proportions to distorted faces and bodies to make them look as normal as those in the center of the image.
 

neuroanatomist

I post too Much on Here!!
CR Pro
Jul 21, 2010
27,555
7,339
i went back to that review and it doesn't seem to be describing what you're saying. and then I went and looked up anamorphosis and it doesn't seem to have to do with the lens.

Here's what it says:

Based on DxO Labs' exclusive geometric correction technology, DxO ViewPoint allows photographers to easily and efficiently restore the natural proportions of the subjects in their images as an integral part of their workflow.

The deformation of subjects located on the edges of photos is due to the use of wide-angle lenses, and affects numerous situations, particularly social photography (e.g., events, marriages) and photo reportage. Thanks to its dedicated tool, DxO ViewPoint easily fixes this flaw, automatically restoring the natural proportions to distorted faces and bodies to make them look as normal as those in the center of the image.
The point I’m making is that correcting for volume anamorphosis introduces barrel-type distortion.

Anamorphosis is a distortion that occurs at the periphery of wide angle shots. It is distinct from the barrel distortion common in UWA lenses. Correction of barrel distortion accentuates anamorphosis (whether that correction is performed by the lens’ optics or with software is irrelevant). The algorithm to correct anamorphosis counteracts the correction of barrel distortion (but it’s not a direct reversal), in fact it 'overcompensates'. In the example posted, the ‘before’ image on the left is before anamorphosis correction, but after correction of barrel distortion. The 'before' image (which is a crop from the left side of an image) has distorted people and straight architectural verticals. The 'after' (with correction for volume anamorphosis applied) has normal looking people but the architectural verticals are curved (as seen with barrel distortion).

If you have DxO with ViewPoint, you can play with the geometric distortion slider and the volume anamorphosis control (diagonal for people) and see the effects.

Here's an example from a family trip to Paris several years ago, before the Notre Dame cathedral burned. Shot is with the EF-M 11-22mm @ 11mm. You can see the barrel-type distortion introduced by the volume anamorphosis correction (especially the columns on the left side). In the crops from the bottom left corner, you can see how the distortion correction makes the anamorphosis worse (subject's face even wider than without distortion correction), and you can see how correcting the volume anamorphosis restores his face to normal, but at the cost of adding barrel distortion across the image.

Anamorphosis.jpg
 
  • Like
Reactions: 1 users

dirtyvu

EOS 90D
Jan 7, 2019
153
115
I see what you're trying to say. But it seems this "volume anamorphosis" term is an invention by DxO.

I do like what that software is doing. While it doesn't seem to quite solve the "distortion" versus "anamorphosis" problem, it's a good compromise. But I'm so entrenched in Lightroom and Topaz Labs especially after the new AI masks that are just godly. But thanks for teaching me a new thing.
 

LogicExtremist

Lux pictor
Sep 26, 2021
501
348
Tldr. What wide angle doesn't have distortion. By its very nature the wider, the more distortion...
The logical fallacy there is that the matter of distortion is a binary state, a qualitative thing, but that's false.
The matter is quantitative,what matters is how much distortion the UWA lens have, not whether it's present in any quantity.

I've already provided comparison figures, the really good UWA lenses I listed have 4% or less rectilinear distortion, which is manageable. Nobody is complaining about the distortion on the RF 15-35 f/2.8 L.

By comparison, the RF 24mm f/2.8 is pushing 11% and the 16mm has 17%, rectilinear distortion, the latter is probably half-way to a fisheye, and both of these lenses aren't able to completely cover a full-frame sensor, so apart from al of the extreme stretching and bending to make the image flat, with software filling in pixels that were never there, there's also a crop required in the software correction to cut off the black corners that the lens couldn't project an image onto, and then an interpolation (software generation) of pixels, much like a digital zoom, to stretch the cropped image back to the camera sensor's full image size which was never captured.
 

LogicExtremist

Lux pictor
Sep 26, 2021
501
348
The point I’m making is that correcting for volume anamorphosis introduces barrel-type distortion.

Anamorphosis is a distortion that occurs at the periphery of wide angle shots. It is distinct from the barrel distortion common in UWA lenses. Correction of barrel distortion accentuates anamorphosis (whether that correction is performed by the lens’ optics or with software is irrelevant). The algorithm to correct anamorphosis counteracts the correction of barrel distortion (but it’s not a direct reversal), in fact it 'overcompensates'. In the example posted, the ‘before’ image on the left is before anamorphosis correction, but after correction of barrel distortion. The 'before' image (which is a crop from the left side of an image) has distorted people and straight architectural verticals. The 'after' (with correction for volume anamorphosis applied) has normal looking people but the architectural verticals are curved (as seen with barrel distortion).

If you have DxO with ViewPoint, you can play with the geometric distortion slider and the volume anamorphosis control (diagonal for people) and see the effects.

Here's an example from a family trip to Paris several years ago, before the Notre Dame cathedral burned. Shot is with the EF-M 11-22mm @ 11mm. You can see the barrel-type distortion introduced by the volume anamorphosis correction (especially the columns on the left side). In the crops from the bottom left corner, you can see how the distortion correction makes the anamorphosis worse (subject's face even wider than without distortion correction), and you can see how correcting the volume anamorphosis restores his face to normal, but at the cost of adding barrel distortion across the image.

View attachment 204732
Using DxO Viewpoint I have found this too, confirming what you've stated.
Thanks for the explanation of how it works, that's really helpful! :)
 

neuroanatomist

I post too Much on Here!!
CR Pro
Jul 21, 2010
27,555
7,339
…there's also a crop required in the software correction to cut off the black corners that the lens couldn't project an image onto, and then an interpolation (software generation) of pixels, much like a digital zoom, to stretch the cropped image back to the camera sensor's full image size which was never captured.
I don’t believe that’s true (although I initially thought that might be the case). Correcting the barrel distortion stretches the corners to fill in the mechanical vignetting. Consider the extreme case of a rectangular projection from a circular fisheye image:

C84E5133-0E88-4786-86D9-64BD7D408A45.jpeg

There’s no upscaling needed after correcting the distortion, the distortion correction itself is doing the upscaling.

Since you have DxO, you can see this effect in less extreme lenses. Because the image is rectangular, barrel distortion correction requires more stretching across the width of the image than for the height. Expand the Distortion control in DxO, there’s a tick box to keep aspect ratio. If that is ticked, you’ll get output cropped to the native sensor size. Untick that box and you’ll get an image that’s the native height but is wider due to the extra stretching needed in that dimension. For example, the Notre Dame image above output without the box ticked is 5275x3456, 91 pixels (~1.8%) wider than the usual 18 MP output of 5184x3456.

That effect is proportional to the amount of distortion in the lens, because more distortion requires more stretching. The EF-M 11-22mm has about 3% distortion at 11mm and is ~1.8% wider after correction. The RF 14-35mm has about 10% distortion at 14mm, and is ~4.5% wider after correction.

So there’s no crop/upscale needed to output native resolution after distortion correction. Rather, the sides of the corrected image actually need to be cropped off to get that native output size.
 
  • Like
Reactions: 1 user

LogicExtremist

Lux pictor
Sep 26, 2021
501
348
I don’t believe that’s true (although I initially thought that might be the case). Correcting the barrel distortion stretches the corners to fill in the mechanical vignetting. Consider the extreme case of a rectangular projection from a circular fisheye image:

View attachment 204758

There’s no upscaling needed after correcting the distortion, the distortion correction itself is doing the upscaling.

Since you have DxO, you can see this effect in less extreme lenses. Because the image is rectangular, barrel distortion correction requires more stretching across the width of the image than for the height. Expand the Distortion control in DxO, there’s a tick box to keep aspect ratio. If that is ticked, you’ll get output cropped to the native sensor size. Untick that box and you’ll get an image that’s the native height but is wider due to the extra stretching needed in that dimension. For example, the Notre Dame image above output without the box ticked is 5275x3456, 91 pixels (~1.8%) wider than the usual 18 MP output of 5184x3456.

That effect is proportional to the amount of distortion in the lens, because more distortion requires more stretching. The EF-M 11-22mm has about 3% distortion at 11mm and is ~1.8% wider after correction. The RF 14-35mm has about 10% distortion at 14mm, and is ~4.5% wider after correction.

So there’s no crop/upscale needed to output native resolution after distortion correction. Rather, the sides of the corrected image actually need to be cropped off to get that native output size.
Thanks, you summed up the critical piece of information when you explained "the distortion correction itself is doing the upscaling".

Got it, so the software correction is taking an image with extreme barrel distortion that doesn't cover the entire sensor, it stretches out the whole image first to a size that would cover the whole sensor (through interpolation), past the position of the mechanical vignetting corners, while compressing the barelled portions of the image flat (presumably by averaging and binning pixels). Due to the barrel distortion, parts of the image (such as the black corners with no image date) get stretched beyond the standard image size, and are cropped away to create an interpolated and corrected image that's the same size as the sensor's standard image. Hope I understood the process correctly! :)
 

neuroanatomist

I post too Much on Here!!
CR Pro
Jul 21, 2010
27,555
7,339
Thanks, you summed up the critical piece of information when you explained "the distortion correction itself is doing the upscaling".

Got it, so the software correction is taking an image with extreme barrel distortion that doesn't cover the entire sensor, it stretches out the whole image first to a size that would cover the whole sensor (through interpolation), past the position of the mechanical vignetting corners, while compressing the barelled portions of the image flat (presumably by averaging and binning pixels). Due to the barrel distortion, parts of the image (such as the black corners with no image date) get stretched beyond the standard image size, and are cropped away to create an interpolated and corrected image that's the same size as the sensor's standard image. Hope I understood the process correctly! :)
Correct. It's important to note that the 'stretching' applies to all distortion correction, not just with lenses having extreme distortion and mechanical vignetting. But for lenses that do, the distortion correction is sufficient to eliminate the mechanical vignetting from the image.

Not that it matters all that much, but in general lens testing is performed without corrections applied. So, for a lens like the RF 15-35/2.8 the test for vignetting shows ~4.5 stops of light lost in the extreme corners. However, after correction of the ~4% barrel distortion those extreme corners are actually cut off, so the real-world vignetting is probably only ~3.5 stops in the corners of the final image.
 
  • Like
Reactions: 1 user
Aug 7, 2018
461
408
It doesn't just seem that way, it IS that way. Canon even said so, explicitly. Check the press release for the RF 16/2.8, where they state, “Long gone are the days of optical corrections…”

Yes, in times of DSLRs with optical viewfinders, you could not make a lens with in-camera or automatic post processing corrections. But with mirrorless, you can, and that is why most (if not all) brands are doing it or will be doing it.

Why? Cheaper and lighter lenses. If the end result is good, I don't care if the corrections are done by using more glass or by using software.
Even for DSLRs corrections were possible in post processing, but there are many good reason to buy a lens that gets rid of all those optical problem with physics instead of processing. Of course I also use lens profile corrections in Lightroom, but they come with a lot of problems. To compensate vignetting for example the corners of an image might have to be amplified a few stops and that includes the noise. It might not be possible to recover shadows in the corner, if the corners were already amplified. The same is true for blurry corners that are corrected by software. You will run into problems if you want to sharpen the whole photo more. While photographers often used Photoshop to "lie" to the viewers, now the cameras use algorithms to lie to the photographer. If you can't even turn off those lens corrections in the viewfinder, that is a major problem.

It shocks me that Canon even seems to be proud that their future lenses let their flaws get corrected by the camera. That pretty much reminds me of the autotune software in the music industry. Your voice can be horrible and you miss most of the tones, but autotune does the job for you. I am sure that with today's computer power that even works in real time. So your concert visitors will never notice that you are a horrible singer, because real time autotune hides that fact.

Those new lenses are not even cheaper than the old ones. The heavy distorted 14-35 is really expensive, but for that price Canon does not even deliver a lens that produces photos with low distortion without the help of software. If Canon continues that route, I might buy a mirrorless camera one day, but only with old EF lenses.
 

neuroanatomist

I post too Much on Here!!
CR Pro
Jul 21, 2010
27,555
7,339
Even for DSLRs corrections were possible in post processing, but there are many good reason to buy a lens that gets rid of all those optical problem with physics instead of processing.
There are also good reasons not to – more correction generally requires more complex designs with more elements, meaning greater weight and higher cost.

While photographers often used Photoshop to "lie" to the viewers, now the cameras use algorithms to lie to the photographer. If you can't even turn off those lens corrections in the viewfinder, that is a major problem.
There are also advantages to seeing the corrected version in the viewfinder. Even lenses 'corrected with physics' aren't perfect. Notably, expensive L-series EF UWA zooms have plenty of barrel distortion and vignetting. Say you frame an image with the EF 11-24/4L at 11mm carefully in a DSLR viewfinder, or in a MILC EVF with the corrections disabled. When you later correct the ~7% barrel distortion in post-processing, you find that the edges of your carefully framed shot have been eliminated, and elements that were visible in the VF are not present in the corrected image. Seeing the corrected version when composing the image can be advantageous.

It shocks me that Canon even seems to be proud that their future lenses let their flaws get corrected by the camera. ... Those new lenses are not even cheaper than the old ones.
If they can offer a lens that costs less to produce but otherwise delivers similar final images, that means more profit for them – I'm sure they are proud of that, although it doesn't benefit the consumer. However, if you factor in inflation between the release of the EF version and the RF replacement, there's not much difference and some RF lenses are actually cheaper. The additional cost is generally getting you more features – wider ultra wide lenses, longer telephoto zooms, smaller/lighter lenses, greater maximum magnification, etc. The one obvious case where the lenses are basically identical is the 24-105/4L, and for that lens the EF MkII and RF versions launched at the same price (and the RF is cheaper with inflation factored in).

In other cases, some lenses are released at prices that are really a bargain. There's no way a 'corrected with physics' 16mm f/2.8 lens would be sold for $300.

The heavy distorted 14-35 is really expensive, but for that price Canon does not even deliver a lens that produces photos with low distortion without the help of software.
I highly doubt a 14-35mm f/4 zoom having low distortion could be designed that would weigh 540 g (the EF 16-35/4 is ~14% heavier than the RF despite being 2mm narrower), and using a 77mm front filter. It's also with noting that by 16mm, the RF 14-35 has ~5% distortion, and at the wide end the EF 16-35/4 has ~4% distortion. So you can sort of view the RF lens as giving you similar performance across the overlapping zoom range, but also giving you an extra 2mm of focal length at the wide end that needs additional corrections, but doesn't cost you anything in terms of weight or size (but does cost you more money).

If Canon continues that route, I might buy a mirrorless camera one day, but only with old EF lenses.
That's your prerogative. IMO, some of the RF lenses have significant advantages over their EF counterparts. Others, like the RF 28-70/2, have no equivalent.
 
  • Like
Reactions: 1 users