Scott Stoness
Offline
Upload & Sell: On
|
 Re: Optical Limits reviewed Canon RF 16mm f/2.8 STM | |
Jesse Evans wrote:
Scott Stoness wrote:
freedolin wrote:
Allow me to comment, especially on our methodology
Scott Stoness wrote:
I really like optical limits however:
1) OL tests uwa by bringing the image in close to the lens until the target fills the page. Thus centre is proportionately way closer than edges. This is a broken method.
No, this is how all MTF tests are done, and there is simply no alternative to it. Interestingly, it's never an issue for lenses with good results, but always a reason to question our testing approach for lenses with fairly low resolution numbers.
Lenses have a flat image field (theoretically). So testing a flat target is the natural approach. For wide angle lenses, yes that brings up the issue that we're not testing at infinity (you can't, really), but if you happen to have a lens with a distance scale (the RF 16 doesn't qualify here), you can still see that with our typical test distance we're still closer to infinity on the lens' scale than to its MFD.
Scott Stoness wrote:
2) Similarly I am only interested in performance at infinity. Which is not tested.
As I wrote already: this is simply not possible. Unless you happen to own a landscape you can fill with a giant MTF test target.
Scott Stoness wrote:
3) They test corrected. The lens is really 14mm at uncorrected. For landscape. The vignetting can be cropped out at 16mm. I will shoot uncorrected and the IQ will be much higher. Most times I crop anyway to 16x9. Unless you are shooting architecture this conclusion is misleading. Uncorrected the iq at corners is much higher (albeit distorted).
We test corrected in this case, and it's clearly stated in the text that results without distortion correction naturally show higher resolution in the image corners. Testing without distortion corection simply would not have been possible due to the massive vignetting. The corners are pitch black, there is nothing to measure there for Imatest. Testing with cropping would have led to results that are simply not comparable to all other RF reviews, since the resolution of the test image would be different.
On top, as also clearly stated in the review: this lens is very obviously designed to be used with software correction enabled. Still, we're interested in the native performance of a lens, since software correction never comes for free. Many manufacturers heavily rely on software correction nowadays, true, but no other manufacturer currently offers lenses that are underdesigned to such a degree as Canon does. I think it's fair to point that out and demonstrate it by numbers.
It's a very personal decision if you can live with the drawbacks of such a lens. And as we also pointed out, it's definitely a price/performance king. On the other hand it's also fair to ask, if the lens could have been significantly better (optically) for a slightly higher price. Personally, I think a $400 oder $450 lens, that under the hood fills the sensor to the corners and is not a hidden fisheye would have sold similarly well.
Markus (OL team)
Obviously I have some passion for this issue or I would not be fussing with it. I spend a lot of $ on lens and if it did not matter, I would just have a $300 lens. So I will say it again clearly:
The review referred to above is based on corrected, not uncorrected. Most of OL tests are done at uncorrected. This is noted in the review. I understand that the reason OL did this is that they view that most people will be using this lens corrected. This is a big deal. I will use the lens uncorrected (but cropped to near 16mm) for landscape whenever possible. And I will miss the high vignetting in corners by cropping.
1) the 16mm f2.8 is really a ~14mm lens with very high vignetting and very high barrel distortion
2) if you crop the raw manually proportional to original dimensions, you remove the vignetting down to a reasonable level (say 3.5%) which is normal for a UWA lens, you get a wider than 16mm lens with relatively higher barrel distortion. eg for example 15.5mm with 3.5% vignetting and 8% barrel distortion.[guessing]
3) Testing the lens in raw based on where you would crop it to in 2) , you would likely find the extremes significantly higher IQ than listed by OL. and quite acceptable. Albeit you are throwing away a small amount of your sensor resolution in cropping.
4) In addition, OL measures at close focus. I do not generally close focus. Its not clear whether closer to infinity focus will be much better or the same or worse, but comparing the lens rentals tests at infinity focus for 12-24 Canon vs the OL tests - the lens rentals shows much better. So I suspect that it will be considerably different.
5) generally all of the above is not a significant for longer mm testing.
I really appreciate OL efforts and consistency. They have a great site that I go to often.
But I note for anyone using OL for UWA, that you really need find a test on a tripod at near infinity centre and edge to find out whether the edges are acceptable for landscape. And perhaps compare it to a similar lens in the same manner.
suggest
The results shown by OL are right if you just leave the auto correct on - which I will do when shooting family shots where I don't much care about the edges anyway.
You're talking about removing ~16.5% of the pixels from your image. I suspect you would not see a significantly different MTF. You'd have to control for comparing a 37MP image instead of a 45MP image, and then you would be left with a near fisheye result.
As for the "close focus" claim, I don't really think that they are using "close focus". The hyperfocal distance for a 16mm f/2.8 is 9.95, infinity is not much beyond that. They are testing much closer to infinity than they are to close focus.
Also, the Gordon Laing video seems to confirm the OL results, rather than refute them. The corners are very poor wide open. He doesn't test the edges (as opposed to corners).
At the end of the day, if the lens works for you, that's great.
I think if you combine the OL and Gordan Laing video and the digitalpicture you would observe:
1) GL: at f8 (near infinity focus) , the 16/2.8 corrected is close to the 14-35f4 at edges
2) GL at f8, (near infinity focus) the 16/2.8 corrected is less than 15-35 f2.8 at edges but comparable [thats pretty good]
3) OL: at f8, (close focussed) the centre is 4700 and the extreme is 2700
4) OL: at corrected, (close focussed) with vignette correction off (cropped to 16mm to remove extraneous stuff) the vignetting is 3.5% [2% at f8]
5) Thedigitalpicture 16/2.8 vs 15-35f4 shows 15-34 [both corrected?] being marginally better at f8 on edges, but this is close focussed to get it in the chart. But 16/2.8 is very acceptable. https://www.the-digital-picture.com/Reviews/ISO-12233-Sample-Crops.aspx?Lens=1573&Camera=1508&Sample=0&FLI=0&API=4&LensComp=1569&CameraComp=1508&SampleComp=0&FLIComp=0&APIComp=3
6) GL: barrel distortion is ~10%
2) 1) is not consistent with 3) - OL edges at 2700 [either infinity focus improves it or beyond 2700 is not bad]
4) 3.5%vignetting at wide open [with dark edges cropped off to 16mm] is not unreasonable for an uwa lens for landscape
6) 10% is pretty distorted but okay for landscape unless you are shooting where there are not many trees/features (ocean, lake, Death Valley)
5) suggests that the 16/2.8 with autocorrection turned off and manually cropped, focussed at infinity, is likely very good. And GL suggests that even with autocorrection, focussed at near infinity, its pretty good - comparable to 15-35f4. With people corrected at f2.8 I suspect that no one would notice either because of bokeh and the attention on the people.
So for $300, its a winner. And even ignoring the price given its weight/sizeand corner performance uncorrected for landscape it quite good. And for hiking at crop mode (19mpx and 24mm) its a great action capture lens (small, light, good resolution).
I will still rather use laowa 15 shift or TS17 shift if I was not hiking far, but for travel or backpacking with 50lbs, this lens plus 24-105 f4-7.1 [and sometimes 100-400 RF] is an excellent kit.
|
