p.17 #2 · Zeiss "Batis" AF lens line to be released for FF E-mount
Matt Grum wrote:
Published MTFs are difficult to compare between the two lenses, as the curves from Zeiss are measured and specified at 10, 20 and 40 lp/mm, whereas Sony's are simulated, don't take into account diffraction and are specified at 10 and 30 lp/mm. At f/2.8 the Sony looks very impressive, the 30 lp/mm line roughly on par with the 20 lp/mm line from the Zeiss, which even taking into account the fact it's simulated vs measured I would put the Sony slightly ahead.
At f/8 the Sony is almost perfectly diffraction limited out to 17mm (86% of the image area), that means it literally couldn't be any better. It's hard to say how much of the Zeiss MTF performance at f/8 is due to diffraction but it's not really possible for the Zeiss to be any better than the Sony.
There are no figures for distortion for the Sony lens but it's hard to imagine it will be anything like 3%. I expect the Sony to be the lens to choose for out and out sharpness wide open and stopped down as well as lack of distortion. This makes it more suitable for landscape and architecture shots. The Zeiss Sonnar will likely have better bokeh, and ought to hold it's own stopped down - it will be fine for landscapes but maybe not cityscapes on account of the distortion.
I'm still torn, as I don't need AF in a macro lens, and want my travel kit as small and light as possible. The Zeiss comes in 125g lighter, which is not going to be massively noticeable, but more importantly 38mm shorter, which means it will go in a smaller bag, look less conspicuous etc. I will probably end up owning both as a fast portrait lens is sometimes necessary, and I will need some form of macro, but can't say which I will get first or which I will bring on trips.
it's nice to some great choices now w/ the FE mount. however, I'm torn b/c I tend to shoot 50/50 between 'central subjects' where I don't care about the edges but prefer nice separation and bokeh and 'landscapes or cityscapes' where I like edge to edge sharpness and limited distortion.
I realize most lenses lean in one direction or the other, so, I just need to find the right balance - probably after looking at hundreds of samples of flickr!
p.17 #3 · Zeiss "Batis" AF lens line to be released for FF E-mount
Matt Grum wrote:
Pincushion is more common than barrel distortion with telephoto lenses. And this is one of those cases where Cambridge in Colour is wrong on something - it isn't taking image size loss / resampling into account.
With barrel distortion, the corners do gain from reduction in area, lose from the resampling process, and both the corners and the centre then lose from the fact the resulting image is too small and must be enlarged to match the original. So that makes for a net loss in both the corners and centre.
With pincushion distortion, the corners lose a lot from being stretched out, and from being resampled in general, but the centre is unaffected as there is no need to enlarge the resulting image, and the centre largely escapes any resampling.
If anyone is having difficulty visualising this, I've made some images.
p.17 #7 · Zeiss "Batis" AF lens line to be released for FF E-mount
Jonas B wrote:
...but we don' always _have_ to resize the image back to the original size. I get lost somewhere there.
You don't, but you do have to crop the image (unless you want big black areas at the sides, or you want to spent time filling in the gaps), which means you've lost resolution. So if you downsample to post online you wont benefit as much from downsampling because you're starting from a smaller image. And if you're upsampling to print you have further to go...
If you were going to crop the image anyway, then you've not lost anything. But if you're always cropping then you're always losing resolution and should think about whether you need a different focal length.
p.17 #8 · Zeiss "Batis" AF lens line to be released for FF E-mount
Tariq Gibran wrote:
That logic makes sense.
Actually I think I've got it backwards, with barrel distortion correction you stretch out the corners leaving the centre intact, and crop off the excess. With pincushion you effectively have to enlarge the centre of the image so it should be much worse. If that's the case then I don't understand the LensRentals blog results.
p.17 #9 · Zeiss "Batis" AF lens line to be released for FF E-mount
Matt Grum wrote:
Firstly pincushion tends to be more common than barrel distortion with telephoto lenses, so I'd be surprised if Lloyd was wrong on this. Secondly, I'm afraid this is one of those cases where Cambridge in Colour is wrong on something - they aren't taking cropping and resizing into account.
With barrel distortion, the corners do gain from reduction in area, lose from the resampling process, and both the corners and the centre then lose from the fact the resulting image is too small and must be enlarged to match the original. So that makes for a net loss in both the corners and centre.
With pincushion distortion, the corners lose a lot from being stretched out, and from being resampled in general, but the centre is unaffected as there is no need to enlarge the resulting image, and the centre largely escapes any resampling. ...Show more →
I think CiC assumes the resampling to be perfect and only considers the area stretching / compression.
The resolution loss due to the resampling operation itself should not be overdramatized. The loss can be quite small, although this may require a large filter order with high memory requirements and a long computation time.
When the distortion requirements are relaxed in the lens design stage, other lens aberrations can be much reduced, and I would rather have to correct distortion in post than any of the other aberrations.
p.17 #11 · Zeiss "Batis" AF lens line to be released for FF E-mount
Right I've done a couple of experiments and got things sorted in my head. Here's an image from the FE 28mm f/2.0 which has significant barrel distortion. On the right is a version I have manually corrected for distortion:
Here is a visualisation of the difference between the corrected and uncorrected version, the black area in the centre shows where the images are the same:
p.17 #12 · Zeiss "Batis" AF lens line to be released for FF E-mount
Matt Grum wrote:
Actually I think I've got it backwards, with barrel distortion correction you stretch out the corners leaving the centre intact, and crop off the excess. With pincushion you effectively have to enlarge the centre of the image so it should be much worse. If that's the case then I don't understand the LensRentals blog results.
I need to do some experiments!
You do essentially crop and then resize when correcting for barrel distortion, a step not needed for pincushion distortion correction. I thought that's what you had initially said.
p.17 #13 · Zeiss "Batis" AF lens line to be released for FF E-mount
Matt Grum wrote:
Right I've done a couple of experiments and got things sorted in my head. Here's an image from the FE 28mm f/2.0 which has significant barrel distortion. On the right is a version I have manually corrected for distortion:
Here is a visualisation of the difference between the corrected and uncorrected version, the black area in the centre shows where the images are the same:
Thus it is possible to correct for barrel distortion without affecting the resolution in the dead centre. However the unaffected area is quite small.
Correcting for pincushion distortion will result in lowering the resolution in the centre, thus it is worse (apologies to Cabridge in Colour)....Show more →
Matt, your "corrected" image is showing an odd distortion that can't be easily corrected. Very evident on the right hand side. It bows inward from top to bottom (more towards the bottom), while the left hand side is "almost" straight from top to bottom. Not a good result for any lens.
p.17 #14 · Zeiss "Batis" AF lens line to be released for FF E-mount
You normally use a biCubic interpolation to do the resampling, and you often choose to give the bicubic quite "safe" parameters.
One thing that you have to understand about MTF is how extremely sensitive to small post-processing contrast changes.
This is why "pre-processed" MTF numbers (imatest et al) are quite useless for real optical performance measurements. Even extremely small amounts of sharpening can make a mediocre lens show the same MTF numbers as a very good lens.
In the same train of thought you should also include the effect of resampling on MTF.
If you resample a VERY GOOD SHARPNESS 3000x2000px image to 2700x1800px with normal bicubic, you will actually lose px-level MTF.
There ARE resampling algorithms that will do the opposite, increase px-level contrast - but they're what you usually call "unsafe". They make stuff LOOK sharp, but the sampling theorem is thrown out the window. The image will no longer accurately model the data underlying the point-sampled image.
I would say that ANY distortion correction in post should be considered to decrease real image performance by about the distortion percent corrected +1.
That is: a 1000px image with 4% distortion should be considered to be a 1000-(1000*(0.04+0.01)) = 950px.
You DO NOT really lose any MTF except for the resampling loss (low) - what you really lose is MP resolution of the underlying optical MTF result.
p.17 #15 · Zeiss "Batis" AF lens line to be released for FF E-mount
Tariq Gibran wrote:
You do essentially crop and then resize when correcting for barrel distortion, a step not needed for pincushion distortion correction. I thought that's what you had initially said.
Do I have this wrong with regard to correction for pincushion distortion somehow?
What I didn't realise at first was there are two ways to perform the correction (that ultimately differ by a scaling factor). To correct for barrel distortion you can stretch the corners out or you can push the edge mid-points in. Likewise to correct pincushion distortion you can push the corners in, or pull the edge mid-points out.
If you correct barrel distortion by stretching the corners out it leaves the centre unrescaled, at this point you are done. If you push the edges in, then you must make the centre smaller, you must also crop the image as there will be gaps (black areas), you must also resize the whole image, the overall effect being the centre winds up being the same size!
If you correct pincushion distortion by pushing the edges out, you end up enlarging the centre and losing resolution there. If you correct pincushion distortion by pushing the corners in, then you have gaps which means you have to crop, enlarge the whole image and thus lose resolution in the centre.
What you've done with your first example is push the edges in, rather than pull them out, if you had done that then neither image would have needed a cropping/resize step, but the pincushion image would still have had it's centre enlarged (by the action of pushing the edge mid-points out.
p.17 #16 · Zeiss "Batis" AF lens line to be released for FF E-mount
Toothwalker wrote:
When the distortion requirements are relaxed in the lens design stage, other lens aberrations can be much reduced, and I would rather have to correct distortion in post than any of the other aberrations.
If it's aberrations that are really hard to remove, like longitudinal CA (axial colour), then I agree with you there, for a fast portrait lens. In a lens used for architecture resolution would be more important to me, so I'd rather a distortion corrected optic.
---------------------------------------------
pdmphoto wrote:
Matt, your "corrected" image is showing an odd distortion that can't be easily corrected. Very evident on the right hand side. It bows inward from top to bottom (more towards the bottom), while the left hand side is "almost" straight from top to bottom. Not a good result for any lens.
That was just a really quick correction to illustrate the point about not touching the image centre, I imagine the proper lens profile would have done a better job. The FE 28 is pretty bad for distortion, though.
p.17 #17 · Zeiss "Batis" AF lens line to be released for FF E-mount
Matt Grum wrote:
You don't, but you do have to crop the image (unless you want big black areas at the sides, or you want to spent time filling in the gaps), which means you've lost resolution. So if you downsample to post online you wont benefit as much from downsampling because you're starting from a smaller image. And if you're upsampling to print you have further to go...
If you were going to crop the image anyway, then you've not lost anything. But if you're always cropping then you're always losing resolution and should think about whether you need a different focal length....Show more →
I'm nearly always losing a little resolution; most of my images end up at an 1:1 or close to that aspect of ratio. For the most of the time I crop a little at all sides of the image. YMMV - anyone composing with the exact final image in mind will lose some pixels.
p.17 #18 · Zeiss "Batis" AF lens line to be released for FF E-mount
Matt Grum wrote:
What I didn't realise at first was there are two ways to perform the correction (that ultimately differ by a scaling factor). To correct for barrel distortion you can stretch the corners out or you can push the edge mid-points in. Likewise to correct pincushion distortion you can push the corners in, or pull the edge mid-points out.
If you correct barrel distortion by stretching the corners out it leaves the centre unrescaled, at this point you are done. If you push the edges in, then you must make the centre smaller, you must also crop the image as there will be gaps (black areas), you must also resize the whole image, the overall effect being the centre winds up being the same size!
If you correct pincushion distortion by pushing the edges out, you end up enlarging the centre and losing resolution there. If you correct pincushion distortion by pushing the corners in, then you have gaps which means you have to crop, enlarge the whole image and thus lose resolution in the centre.
What you've done with your first example is push the edges in, rather than pull them out, if you had done that then neither image would have needed a cropping/resize step, but the pincushion image would still have had it's centre enlarged (by the action of pushing the edge mid-points out.
The first example I have shown is essentially what ACR/Lightroom does behind the scenes (or exactly what you see if you perform the actions manually in ACR/Lightroom). I don't think there is a way to avoid the resize part with barrel distortion. Even if you don't think of it as a separate step, by "pulling" on those corners as you suggest, you would essentially be interpolating up in size the entire image (except for perhaps the exact center). Conceptually, it probably doesn't matter how you think of it occurring as the end result (resizing of most of the image) still occurs. I think we are kind of going in circles and confusing ourselves at this point.
p.17 #19 · Zeiss "Batis" AF lens line to be released for FF E-mount
There are some sample images up on Flickr from the new Batis lenses. Plus one of the images has dimensions that belongs to none of the current a7 series cameras.
p.17 #20 · Zeiss "Batis" AF lens line to be released for FF E-mount
jsv_20 wrote:
There are some sample images up on Flickr from the new Batis lenses. Plus one of the images has dimensions that belongs to none of the current a7 series cameras.
they must deliver the newspaper a few days late in Ohio as well......
