More 16 F2.8 comparisons: corrected vs. uncorrected vs. JPG

Sittatunga wrote:

There's been a lot of discussion about this, including another thread here; https://www.dpreview.com/forums/post/65558766 which was about the effect of this lens on a 20MP R6

The geometry of what happens when you correct barrel distortion is quite complex. This is what happens, step by step. I don't have the 16mm lens yet, but I do have a compact 20MP camera with a10.4-37.1mm zoom which has a very similar level of barrel distortion at the wide end, though not quite as complex. First, I photographed a brick wall to match as near as possible the one used in the thread I linked to. I know it's a terrible brick wall for the purpose, but that doesn't really matter for what I'm trying to show. Then I opened the RAW file in the Windows 10 Photos app and saved it as a jpeg, this one.

20 MP, 5472x3648 uncorrected.

Next I opened it in the GIMP and increased the canvas size to 9000x6000 pixels so we can see what happens to the corners and edges when we adjust the distortion. I also added some 100x100 pixel black squares near where I though the corners would come and somewhere near the middle of the sides and image. Then I used the GIMP's Distorts/Lens Distortion... filter to correct the distortion to near enough the out of camera and PhotoLab corrected jpegs and cropped to the acute corners.

This is a 1920x1280 resample of the 7287x4857, 19.7 MB jpeg I saved from the GIMP

7287:4587 is an aspect ratio of 1.59:1. That's because barrel distortion is a decrease in image scale varying with distance from the centre. Correcting it increases the original image scale, again varying with distance from the centre. The short ends of the image are further from the centre, so they will get magnified more, so the aspect ratio increases. The original 10000 pixel square in the bottom left corner has grown to a lozenge covering 93 314 pixels. The longer diagonal has grown by 3.56x and the shorter has grown by 2.75x, which is why the corner resolution goes down.

Next, I cropped to the largest rectangle I could fit inside the image, 6276x3832 (1.64:1 aspect ratio) which is still 20% bigger than the 5472x3648 original. Maybe DPP and PhotoLab do a little re-sampling to get the shorter dimension to equal the original, I couldn't say for sure.

Resampled to 1920x1172

Finally I cropped it to the 1.5:1 aspect ratio everybody's been arguing about.

5748x3832

PhotoLab's version for comparison, with added clarity and noise reduction

5472x3648

Not exactly apples to apples but it gave me some idea of what actually happens in post-processing lens distortion correction. Of course, you lose quite a bit of resolution at the extreme corners, but what lens doesn't?

Higgins2002 wrote:

PAntunes wrote:

sportyaccordy wrote:

Alastair Norcross wrote:

sportyaccordy wrote:

Alastair Norcross wrote:

sportyaccordy wrote:

This lens is a mixed bag for me

Good price, FL and spec.......... but corrections rob you of a good bit of resolution. Just guessing it looks like a 1.1-1.2x crop which is like 20-40% of your MPs.

As far as I can tell, applying the corrections I used is equivalent to about a 1.15 crop, which loses about 25% of your MP.

For a lens that I imagine will be used for landscapes a lot that's kind of a bummer

That would be more of a problem in the days of 8MP sensors. But even then it wouldn't have been terrible. I still have my 8MP 20D, which was an amazing camera in its day, and gave me lots of great landscape shots. This lens would have used 6 of those 8. My first DSLR was the 6MP 300D. I have several 30" X 20" poster prints of landscapes from the 300D on my walls. They look great. From a 6MP sensor. Now that we have megapixels oozing from every pore, it's even less of a problem. My R uses about 22.5 of its 30 megapixels to produce the corrected images from this lens. I can remember when everyone was waxing lyrical about the incredibly high resolution of the 1DsII, which had 17 MP. If you use an R5, you'll still be using about 33.5MP with this lens. I know that people (on these forums at least) keep insisting that they 'need' more and more megapixels. But they really don't. They're just trying to justify their lust for newer and newer gear. I prefer to not attempt to rationalize that lust. I just admit that I really like new toys (not that I can usually afford them). Need doesn't come into it. It's all about want.

But I guess that's reasonable to expect for the price/size. Kind of seems like a missed opportunity like the RF 50 1.8

Im not following your logic. If more MPs dont matter youd still be shooting a 6MP camera.

Plus I'm not even suggesting that people run out and buy super high MP bodies. But there are photography disciplines that leverage resolution. A lens that throws out 25% of the MPs it captures will be less desirable to a lens that doesnt, regardless of how many MPs are available.

other things being equal, which they never are. My point, which I didn’t think was that difficult to follow, was that using 75% of the pixels on a modern camera still gives great results. More pixels provide diminishing returns.

You dont get to decide this for everybody. For you it does. For some maximizing resolution is important. Neither approach is right or wrong, but having info like the amount of correction the lens needs is useful in making that decision.

For many people, a small, light, cheap lens that uses 75% of the massive number of pixels we have available is a far better proposition than a big, heavy, expensive lens that uses all of them.

Having that info allows people to choose the lens that makes the most of their sensors if that is a priority.

This lens does raise a good question... What happens in software with the correction?
Does DPP maintain the original resolution (aka upsamples the entire image) or does it actually remove the resolution, eventually creating a problem with lower res cameras?

The original captured image is for R5 8192 x 5464.

If you cut off those corners and stretch it too be straight you loose pixels for sure, so if the final picture is 8192 x 5464 after corrections they had it upsampled.

But if you see the original examples, even the top and bottom parts of the image are cropped. Quite easy to see on the first image with the space above the window in the middle.