Home · Register · Software · Software · Join Upload & Sell

  

  Previous versions of Ayoh's message #14172450 « Jordan at Admiring Light reviews Sony FE 70-200f/2.8 GM »

  

Ayoh
Offline
Upload & Sell: Off
Re: Jordan at Admiring Light reviews Sony FE 70-200f/2.8 GM


Brandon Dube wrote:
Ayoh wrote:
Brandon Dube wrote:
Ayoh wrote:
RCicala wrote:
Oh, I do despise testing this lens. But because I get so misquoted about it, let me throw out a couple of things.

1) It's about as good as the Canon 70-200 f/28 IS II at 70mm and 135mm. At 200mm it's not quite as sharp in the center, and has more copy-to-copy variation. But it's not awful, it's still good.

2) Neither of the above are nearly as good as the Nikon ED. Nothing else is either.

3) The Sony has higher sample variation at 200mm than most other 70-200 f/2.8s, including the Canon, which itself isn't that great at 200mm as far as sample variation.

4) All our results are at infinity and without a camera. DxO and Imatest results test it at closer focusing distances. But they also test with a camera. If you believe (like I do) that Sony cameras are better then it stands to reason that the Sony system would test better, since they're testing the system, not just the lens. I think this explains a lot of why DxO finds every new Sony lens "The best we've ever tested

We've never tested any lens as thoroughly as we've tested the Sony 70-200 f/2.8 GM. Why? Because Sony said we had to be wrong. We weren't; they were. By the time we had proved that to my satisfaction I was so sick of that lens I wished I'd never seen one. This colors my comments, I put it out there because I always want to identify when I might not be impartial.


Roger, based on your comments here you should update the post linked below as it has inconsistent conclusions.

https://www.lensrentals.com/blog/2017/01/an-update-and-comparison-of-the-sony-fe-70-200mm-f2-8-gm-oss/

Based on the published out-of-camera resolution chart images the Sony is either misaligned or decentered (as the left side is softer than the right) but despite this appears to be sharper on the right side than both the Canon and Nikon at both 70mm and 200mm. For example, below are the bottom right corners of the published images.

https://imgur.com/a/RrU6H

This may just be a fluke due to sloppy test chart photos or it may suggest that lens is capable of higher sharpness on camera than on your rig. Or it may be that the Sony performs better at close focus distances than infinity and hence the discrepancy between your rig measurements and the test chart measurements (and the results from DXO etc).
The misalignment in the Sony also suggests that the lens is certainly subject to inconsistency and variation.



If you take something that is very inconsistent, it could be very good or very poor. The average turns out somewhere in the middle.

We average over the full 2D field to generate the 1D MTF vs Field. So yes, some 70-200 GMs are great on the right side. They're probably garbage on the left side. Maybe the copies of the Canon and Nikon lenses Joey picked up were better aligned at 200mm than the Sony was.

With a 70-200mm f/2.8 lens, it is almost random chance.

The MTF bench is orders of magnitude more well controlled than a chart test. It will not accrue substantial error until past 1,000 lp/mm.


Ok thanks, so the reported MTF measurements for a single lens are averages at multiple orientations (and poor uniformity brings the MTF down for that lens). But are the deviation bands shown in your variation plots (linked example below) the deviation in the lens-averaged MTF, or the deviation of all your individual MTF data points? i.e. is it analogous to the standard deviation of all your individual MTF data points across many lenses, or the standard error in the averaged MTFs for each lens?







I guess it is the standard error since the maximum values in the 50LP/mm variation bands for the Canon lens above, are higher than the corresponding max value for the Sony lens below, even though we know that at least one of the Sony lenses in your stock is much sharper than the Canon on one side (based on the images referenced here http://www.fredmiranda.com/forum/topic/1507053/2#14170778)







So I guess your deviation bands will not show the maximum MTF values that were achieved by any one lens at a specific orientation and therefore not reflect the maximum observed performance (which is fine, it makes sense you had to do some averaging). I guess Sony needs to bring the variation down to reflect the potential achievable best in class performance in your charts (as seen on the right side of the lens tested here (http://www.fredmiranda.com/forum/topic/1507053/2#14170778). I think I really did get lucky with the two excellent copies of this lens I tested.



Drawing conclusions about a population based on an observation of a single member is kind of silly. The fact that the particular lens Joey did his test with is good on the right hand side doesn't really tell you anything about the lens. If it is bad on the left and good on the right, it is good on the right because it is bad on the left, and a perfectly aligned one might be "meh" or "okay" on the right, but won't be great.

If we computed the average charts in a way that was indicative of maximum performance, that would be very dishonest.

The aggregation is done based on the assumption that the lenses should all be rotationally symmetric. Per lens, we get 8 data points for 2mm .. 20mm. We take these vectors of length 8, and concatenate them all together for each lens and are left with an N*8x1 vector for each field point, where N is the number of lens copies tested. We then average and compute stdev over the N*8 dimension.


You missed my point. To reiterate:
1) at the moment your variation bands correspond to the standard error in the means of the MTF values averaged for each particular lens
2) the variation bands could be set so they correspond to the standard deviation of the total set of all unaveraged MTF values recorded for all data points across all lenses

For scenario 2 the maximum and minimum values of the variation bands would have higher magnitudes and would be a closer indication of the max/min observed MTF values. As you plot case 1) the max/min values are reduced due to averaging. Therefore high MTF outliers like ones for the right side of the particular Sony lens above are not captured in the current approach. This is not necessarily a criticism of your approach, just an observation that due to the averaging the various bands are narrower.



Sep 07, 2017 at 03:20 PM





  Previous versions of Ayoh's message #14172450 « Jordan at Admiring Light reviews Sony FE 70-200f/2.8 GM »