• Members 796 posts
    Nov. 25, 2023, 6:33 p.m.

    I quess most of us read lots of reviews concerning gear we are thinking about buying. Part of the fun is trying to sort the honest reviews from those who have received a bung of some sort. After about a year of indecision, I finally pressed the buy button on a lens that I hope will move my photography forward. A 20% discount helped!

    The seemingly impartial reviews I read could be basically divided into two camps. On the one hand there were those based on just real life usage (Northlight and Philip Reeve and Amateur Photographer) which were pretty favourable, and those that were more "scientific" with MFT tests and such (Camera labs) which were a bit off-putting.

    But after reading too much, I find the only way to test a new piece of equipment, and to establish if it is any good, is to use it in my real life shooting situations. I took my new Laowa 15mm Zero D shift lens out to a place near home, and shot some tall buildings. I shot hand held at high ISO, but my preliminary findings, mostly line up pretty well with the reviews. This lens is no sharpness monster as the technical reviews pointed out, but Camera labs are very wrong when they say the lens is pointless if you own a 14-35 and correct in post.

    But as always my own testing throws up some interesting and useful information about a new piece of kit.

    How do you, if at all evaluate a new piece of equipment?

    DSC_4599 1.jpg
    Zero correction

    DSC_4594 1.jpg

    To fill the frame I was surprised how close I had to get to the building to fill the frame. One thing only personal handling can show you.

    DSC_4615 1.jpg

    DSC_4575 1.jpg

    Branches are merciless exposers of defects.

    Interestingly I tried to correct the uncorrected tower block shot. Capture 1 will not correct beyond a certain point and DXO Viewpoint 1 needed shrinking vertically to see plausible results. ( Maye a later version or other software can do better).

    DSC_4597 1.jpg
    Capture 1

    DSC_4599_DxOVP.jpg
    DXO

    DSC_4599_DxOVP.jpg

    JPG, 1.1 MB, uploaded by NCV on Nov. 25, 2023.

    DSC_4597 1.jpg

    JPG, 653.1 KB, uploaded by NCV on Nov. 25, 2023.

    DSC_4575 1.jpg

    JPG, 1.4 MB, uploaded by NCV on Nov. 25, 2023.

    DSC_4615 1.jpg

    JPG, 1.6 MB, uploaded by NCV on Nov. 25, 2023.

    DSC_4594 1.jpg

    JPG, 926.7 KB, uploaded by NCV on Nov. 25, 2023.

    DSC_4599 1.jpg

    JPG, 900.2 KB, uploaded by NCV on Nov. 25, 2023.

  • Members 1070 posts
    Nov. 25, 2023, 8:49 p.m.

    NCV,

    It sounds like you have the same shift lens as I have. I am generally very happy with the 15mm f/4.5 shift lens. I find it sharper than my Laowa 12mm Zero D f/2.5 and both of them benefit from Topaz Sharpen AI. On the tripod I use it with the mounting bracket.

    As to how I test lenses, I do it by taking photos of real live things as you have done. I dont have any test rigs, but I do take photos of the 2.5m tall bookcase 3 m away from me, before venturing outside.

    It is not clear to me which of your shots have used shift, and I am surprised that you have achieved such strange looking results.

    To check whether the proportions of the final result are correct, it helps to take a non shifted photo from a distance and compare. I use DXO’s PhotoLab 6 with their Viewpoint plug-in for raw processing and final adjustments, and have no complaints.

    David

  • Nov. 25, 2023, 9:12 p.m.

    I did start off by doing the brick wall test. But 2 things irritated me about that:

    1. I don't take pictures of brick walls, so what relevance is it to me
    2. At wide apertures, the edges were getting soft.I then realised that, of course they would - the edge is actually further away from the lens than the centre so at wide apertures, it was outside the focussed depth of field. I would need a curved brick wall.

    So, I stopped doing that and just took pictures of what I normally take. Then look at the results - if I was happy, the lens stayed. if not, it went back.

    ALan

  • Members 796 posts
    Nov. 25, 2023, 9:34 p.m.

    Pictures 2,3 and 4 were taken with this lens which is the same as you own. They are all shifted. The tower block is at full shift. I was no too interested in composition, I wanted to see things like sharpness and CA.

    One interesting thing arose. It seems at full vertical shift, none of the software I own is capable of correcting (picture 1) keystoning properly. So already I am happy I have found a solution for tight spaces. I was just 15 meters away from the tower block, it was a bit mind blowing to get a shot like this with everything in the frame.

  • Members 796 posts
    Nov. 25, 2023, 9:38 p.m.

    Good point (2) about using brick walls. Camera labs actually do a test at infinity, which is pretty useful in my mind.

  • Members 272 posts
    Nov. 25, 2023, 11:48 p.m.

    Every lens has field curvature. The focal plane of a lens is not really a plane, starts out as a surface of a sphere and lens designed add more and more glass to try to flatten it. Brick walls are planes. If you are 10 feet from a brick wall with a lens with 90 deg angle of view - the distance to the brick wall in the edge of the frame is 14 feet or 40% greater than to focal distance of the lens. If it is focused at 100 ft, more than likely the brick wall is not long enough to cover the entire frame and the edge of the wall is well within the DOF. In the lensrentals.com blog there is an interesting post on the testing they did on field curvature.

    On the other hand lens tested only at infinity is not all that realistic unless one only wants to use the lens at infinity.

  • Members 706 posts
    Nov. 26, 2023, 3:54 a.m.

    You can drive yourself totally bonkers doing home lens testing. So many variables, that don't show up in your average amateur test scenarios. As mentioned above, field curvature that you won't really notice in real world 3D scenes, but may look terrible shooting brick walls & test charts etc.
    I prefer buying the lens, then going out shooting with it in my typical use case, if it produces good results for the most part then I'm happy. The issue with this, is that once you've decided, then you're way outside any return window, & the lens will more than likely look used/second hand. Here in Australia, there is no change of mind return policy, once you've parted with your hard earned, you own it. For good. Unless it's obviously broken or defective. I can understand why, as I've read plenty of posts on interwebz forums of certain people going through 1/2 a dozen lenses, before finding a "good" one to keep. The harder you go looking for problems, the more you'll find. As they say, ignorance is bliss ha ha ha ha ha.

  • Members 796 posts
    Nov. 26, 2023, 8 a.m.

    That is it. The only way to judge if a lens is good for your usage, is to actually use it for the subjects you like to shoot.

    I have come to the conclusion that MFT and all the other tests that make most lenses look horrid, are of limited value. They miss a lot of things like micro contrast, rendering and such. I briefly owned a Samyang 24TS, thinking this cheap version of the Nikon TS, might be sufficient. The reviews were quite good. But in actual use I hated this lens. awkward to use with cut price movement controls and a soft low contrast rendering that i hated.

    In Italy too, we do not have the possibility of returning stuff we do not like or which we just want to try out. At least I know when I buy online I am getting something that sombody else has not tried out. The America "returns" culture is pretty crazy.

  • Members 796 posts
    Nov. 28, 2023, 9:56 p.m.

    Well, I tested out my new Laowa 15mm Zero D shift lens seriously on Sunday. Anchored to a good heavy tripod, IBIS switched off, this lens is more impressive in real life use, more than most of the reviews led me to believe. It helped that I got a copy with little decentring.

    This was the review that came closest to my first experience.

    But it was a guy doing a YouTube demo/review shown as a link, on the Italian resellers site that clinched it. He shot some stuff, like I like to shoot, at a place I know, so I was able to evaluate the usefulness of the lens for my photography.

    DSC_4737 1.jpg

    DSC_4636_HDR 2.jpg

    Wide open at full shift the edges fall apart as expected, in fact they are terrible, as these lenses are designed to be shot well closed down. But closed down to F8 as in the second shot the edges at full shift are very good. ( Look at the crosses on the pinnacles in the first shot at full size)

    DSC_4704 1.jpg

    Finally using a solid tripod makes a huge difference to the final image, perhaps more than lens defects.

    DSC_4704 1.jpg

    JPG, 826.0 KB, uploaded by NCV on Nov. 28, 2023.

    DSC_4636_HDR 2.jpg

    JPG, 904.9 KB, uploaded by NCV on Nov. 28, 2023.

    DSC_4737 1.jpg

    JPG, 1003.6 KB, uploaded by NCV on Nov. 28, 2023.

  • Members 692 posts
    Nov. 29, 2023, 11:18 a.m.

    It only seems that way when you are reading some posts. If you return too much stuff, Amazon, bhphoto, etc., would ban you. On the other hand, with online sales, you have no idea what you are buying. There should be some protection for you. Also, is a decentered lens defective? How do you prove that? What if it is softer than the online tests you have seen?

    The merchants do not allow this out of the goodness of their hearts. This policy creates more profits eventually. As a costumer, I am more likely to buy an item if I know that am not stuck with it even if it is junk.

  • Members 1070 posts
    Nov. 29, 2023, 12:17 p.m.

    Yes. I am also very pleased with my copy of this lens (Canon RF version). I find that the lack of sharpness can easily be fixed in Topaz Sharpen AI.

    David

  • Members 796 posts
    Nov. 29, 2023, 4:11 p.m.

    I found the sharpness with the lens stopped down to F8/11 is pretty good without much further PP for my usage. I believe the filter stack on the sensor affects the sharpness of this lens at the edges, form what I have read.

    This lens comes somewhere between my excellent modern Nikon PC lenses and the old 28mm and 35mm warhorses. I saw a test where it was better on the edges at full shift than the Canon 17mm TS

  • Members 796 posts
    Nov. 29, 2023, 4:17 p.m.

    Quite true, but a decentred lens is defective, so it need returning anyway. Or if you are buying second hand if the object does not match up with the description, then consumer rights come into play.

    But I was referring to those multitude of posts on DPR where people buy stuff just to try it out and then return it. It is good that they get put on a blacklist. Because there are lots of forum posts where people received what looks like pre owned gear, sold as new.

    I used only to buy locally, but increasingly I buy online. So far so good, even with second hand gear.

  • Members 692 posts
    Nov. 30, 2023, 12:19 p.m.

    For proper correction, the SW needs to know the FL or some relevant info about it. Say you used a very long lens, which basically means you take a small crop of what you have done. Then you still need to correct the converging verticals by the same angle but the vertical stretch would be more or less uniform over the image. With a UWA, the top should be stretched much more. Maybe the problems is that the SW cannot get the FL from the EXIF.

  • Members 272 posts
    Nov. 30, 2023, 2:54 p.m.

    [/quote]

    For proper correction, the SW needs to know the FL or some relevant info about it. Say you used a very long lens, which basically means you take a small crop of what you have done. Then you still need to correct the converging verticals by the same angle but the vertical stretch would be more or less uniform over the image. With a UWA, the top should be stretched much more. Maybe the problems is that the SW cannot get the FL from the EXIF.
    [/quote]

    Yep. It all has to do with angle of view. With a longer lens, the keystone distortion is close to linear. So knowing the focal length will allow more accurate correction of geometric distortion. It won't be perfect but it will be close enough. However, with a UWA, the the nonlinearity of the keystone distortion is large and even with the focal length only correcting for the focal plane will leave significant residual distortion on planes behind the focal plane. Also UWA lenses have more field curvature than than long lenses not only because of the optics but also the fact the focal distance tends to be closer on a WUA and the closer the focal point the greater the curvature of the sphere defined by the focal distance compared to a longer lens. So while S/W correction might be sufficient for lenses "normal" and longer. However, for WA the S/W changing the lens geometry (tilt/shift) will give better results.

  • Members 1070 posts
    Nov. 30, 2023, 3:40 p.m.

    I find this very unclear. Do you have any references for these statements? Is there somewhere I can read about it and derive the necessary equations for correction? Are you referring to shifted lenses, or a tilted camera/lens?

    Thanks!

    David

  • Members 796 posts
    Nov. 30, 2023, 9:04 p.m.

    In viewpoint I gave the test image as much vertical anamorphoses as the program could provide.

    DSC_4599_DxOVP_DxOVP.jpg

    Looks a bit better with the keystoning sorted in Capture 1 and then corrected for stretching in Viewpoint. But my version of viewpoint has big problems with the original image

    Using the quick "vertical lines" on the original image, Keystone goes mad.

    DSC_4599_DxOVPoooo.jpg

    Using the sliders, it is on the way to being similar to the shift lens image. I adjusted as much as the program will allow, using the slider panel.

    DSC_4599_DxOVPxxx.jpg

    Looks like all these programs have limits to the algorithm, when faced with very high level keystoning. The 15mm PC goes beyond where the two software solutions I have to hand, can arrive.

    DSC_4599_DxOVPxxx.jpg

    JPG, 1.0 MB, uploaded by NCV on Nov. 30, 2023.

    DSC_4599_DxOVPoooo.jpg

    JPG, 957.2 KB, uploaded by NCV on Nov. 30, 2023.

    DSC_4599_DxOVP_DxOVP.jpg

    JPG, 883.0 KB, uploaded by NCV on Nov. 30, 2023.

  • Members 272 posts
    Nov. 30, 2023, 9:38 p.m.

    I one could take an image from a camera and correct all the distortion in S/W, then one would be able to recreate the 3-dimensional image. As we all know, it requires two images of the same subject from differing points of perspective in order to do that. That's why most animals have two eyes. To reconstruct a 3-dimensional space from a single image is what is called in mathematics an ill-posed problem. That is a fancy way of saying there does not exist a unique well defined solution.

    www.collimator.ai/reference-guides/what-is-an-ill-posed-problem

    As for references to 3-dimensional reconstruction, consult just about any text on computer vision. The distortion correction on multiple planes in space projected onto an image plane is also an ill-posed problem,

    cvg.cit.tum.de/research/image-based_3d_reconstruction/multiviewreconstruction#:~:text=3D%20Reconstruction%20from%20multiple%20views&text=Usually%20the%20camera%20position%20and,a%20pixel%2Dwise%20correspondence%20problem.

    en.wikipedia.org/wiki/Computer_stereo_vision

  • Members 796 posts
    Nov. 30, 2023, 9:55 p.m.

    Maybe it is just easier to dig out a shift lens in the ambit of this conversation!

  • Members 692 posts
    Nov. 30, 2023, 10:17 p.m.

    Repeating it multiple times doesn't make it right. It was wrong before, it is wrong now.

    Except that in our case, it does.

    The "fancy" way to say that there is no unique solution is to call the problem underdetermined. "Ill-posed" is typically reserved for problems where the solution is very sensitive to the data, roughly speaking. Typical examples are a deconvolution with a smooth kernel or the small divisor problem. In our case, it is a well poser problem.

  • Members 692 posts
    Dec. 1, 2023, 3:20 a.m.

    Here is the DXO ViewPoint v.3 version from the not-tilted shot:

    dsc-4599-1_DxOVP.jpg

    It complains that there is no EXIF; this is in manual, maximal correction. They limited it there but you can save it and run it again!

    The view points are different though.

    dsc-4599-1_DxOVP.jpg

    JPG, 497.8 KB, uploaded by JACS on Dec. 1, 2023.

  • Members 581 posts
    Dec. 1, 2023, 4:01 p.m.

    Ref 2. Probably someone mentioned it already, but if you have a reasonably good lens and you focus the middle part of the frame at a brick wall, then the edges of the frame "should" also be in focus despite them being further away. This is because the focal plane is designed by the manufacurers to be as flat as possible.
    the reality is that complete flatness is not achieved perfectly, the focal plane can be imagined like an undulating surface with slight bumps and waves and ripples. but it's almost flat and perpendicular.
    It's good that it is flat like that, for example when you are taking a photo of a group of people all standing in a row (perpendicular to the line of sight) they are all in focus, it would be counter productive to arrange the group in an arc.
    This wavey nature of the focal plane becomes more noticable when the lens is wide open and the DOF is very narrow As you stop down, the width of the focal plane increases and there is more leaway to compensate for those undulating imperfections in the field curvature.

  • Members 272 posts
    Dec. 3, 2023, 3:23 p.m.

    Repeating it multiple times doesn't make it right. It was wrong before, it is wrong now.
    [quote]

    In radar imaging, synthetic aperture radar, the image of any plane can be recovered through focusing the phase history data after capture. In magnetic resonance imaging the same can be accomplished. The reason is the captured image data contains information of three dimensions. The data from a SAR or MRI are not images but captured data of three space. Once the image formation is performed and the phase history data discarded, one cannot modify projections of object out side the selected focal plane since the information does not exist in the formed image. Both imaging technologies function by taking multiple planer images of the subject using the geometric relationship between objects in the scene and the position of the sensor to process the data and form the projection onto the selected image plane.

    In order to do that with optical images requires a plenoptic camera, a.k.a., light field camera. Plenoptic cameras record three dimensional information which can be focused on a chosen plane after the capture. In the process of image formation, just as in SAR or MRI, the three dimensional data can be manipulated in order to make all lines parallel if one choses. Once the planer image has been generated, that is no longer possible since the three dimensional information no longer exist.

    en.wikipedia.org/wiki/Light_field_camera#Standard_plenoptic_camera
    www.plenoptic.info

    Granted for lenses normal or longer on a sensor, most of the keystone correction S/W does a reasonable job. Capture One does fine. The Nik plug in Perspective does okay. The DXO application does okay. However, if one were called up on to accurately measure the dimensions of an aircraft on the tarmac an image that was sitting at an angle in three space across the focal plane in both elevation and azimuth, then one needs either multi image stereo imaging if optical images are used or the recorded data from a SAR of plenoptic camera. The act performing a projective transform as our cameras do, destroys metric geometric information That is a projective transform is not a rigid motion - it does not preserve length or angles.

  • Members 692 posts
    Dec. 3, 2023, 4:05 p.m.

    Please, stop it. I actually work in those and related fields. SAR data is 2D. Tall buildings look collapsed and tilted towards the receiver but they are usually plotted inverted. Also, you see multiple images of the same object because of multiple reflections. MRI is completely different than SAR. Now, with multiple antennas, you get something like "stereo SAR." This is what you read on wikipedia. Most of the time SAR is 2D though.

    Here is a typical SAR image, taken from here www.capellaspace.com/data/sar-imagery-products/:

    www.capellaspace.com/wp-content/uploads/2021/01/CAPELLA_C02_SP_GEO_HH_roswell_header.jpg

    Absolutely wrong for SAR. It does not take planar images at all. And I do not even want to start with MRI.

    But... this is not what the keystone correction does. It is not trying to measure accurately anything. It is mapping one projection onto another, both 2D. You do not need the 3D image for that.