[Klaus]

Hi folks,

 

it seems as if the latest Canon review (actually also the others) caused some stir out there.

The issue is that at 50mp the lenses are scaling in the center but at a much lesser degree towards the borders - thus causing an extreme spread between the center and border results. 

 

Thus we are considering to move to a "near center" approach rather than "dead center" (although it never really was the "dead center" - just darn close to it).

"near center" is 1/3 of the distance towards the sides.

 

See the attachment

top: chart as is (for the Canon 35mm f/1.4 USM L II)

middle: near center approach, same scale

bottom: near center approach, adjusted scale

 

Thoughts ? I actually quite like it.

 

[Brightcolours]

You can't win here. The near center approach puts the 5DS-R in a disadvantage in the "center" measurement compared to other cameras, skewing the results another way. It makes "superficial" comparisons of lenses on different bodies still a nightmare?

 

And the border/edge results are hardly lifted in the graph, you still will need a reference test with a lens tested on the D800 and 5D mk II...

[davidmanze]

 Forgive me but could you accurately describe the use of "scaling"  and how you are applying it Klaus? although I have an idea of the principle a bit of clarity would be useful , I'm sure this could also help others too..........

[Klaus]

You can't win here. The near center approach puts the 5DS-R in a disadvantage in the "center" measurement compared to other cameras, skewing the results another way. It makes "superficial" comparisons of lenses on different bodies still a nightmare?

 

And the border/edge results are hardly lifted in the graph, you still will need a reference test with a lens tested on the D800 and 5D mk II...

 

Sorry, the tests were never meant for camera comparisons. Whether the 5Ds R looks bad compared to a D810 is completely irrelevant in my book.

 

Besides - once we switch to the next Nikon (and Sony) we'd follow the same approach. 

 

The near-center approach would also favor higher quality lenses which would be a good thing.

[Klaus]

 Forgive me but could you accurately describe the use of "scaling"  and how you are applying it Klaus? although I have an idea of the principle a bit of clarity would be useful , I'm sure this could also help others too..........

 

Well, the ceiling of the chart is roughly defined by a "neutral" sharpening Imatest figure using a high end lens at an optimal setting. This approach is the same across all systems and invariant of the AA filter strength. The tuning factor is the base sharpening in the RAW converter. The idea is that the sharpening perception is the same on pixel level.

The bottom end is roughly defined as 1/3 of the ceiling but adjusted by a correction factor based on the sharpening used to achieve the mentioned neutral setting. From there on the scale is divided into the 5 segments representing the "school marks" or our version of the "subjective quality metric" if you like. 

 

This also means that technological progress on the sensor side is pushing all those values upwards making it more difficult for the lenses to achieve top marks. That is intentional. If the manufacturers release megapixel monsters, they have to push their lens quality as well in our opinion. 

 

Another way how to look at it:

On the 5Ds R the poor rating starts at a local lens resolution equivalent to the density of an 8 megapixel sensor.

[davidmanze]

OK Klaus, thanks for that!

[Rover]

Howsabout doing both...  But in any case, "near center" might be misleading even for "rough" comparisons with other systems. Same with the logarithmic scale you proposed elsewhere... By the way, my personal gripe: I was never thrilled by the terms you used to name the parts of the scale where "Good" actually meant quite soft results (not sharp enough for critical use, at the very least).

[joachim]

Hi folks,

 

it seems as if the latest Canon review (actually also the others) caused some stir out there.

The issue is that at 50mp the lenses are scaling in the center but at a much lesser degree towards the borders - thus causing an extreme spread between the center and border results. 

 

Thus we are considering to move to a "near center" approach rather than "dead center" (although it never really was the "dead center" - just darn close to it).

"near center" is 1/3 of the distance towards the sides.

 

See the attachment

top: chart as is (for the Canon 35mm f/1.4 USM L II)

middle: near center approach, same scale

bottom: near center approach, adjusted scale

 

Thoughts ? I actually quite like it.

 

There is nothing in my view, which prevents publishing all four bars - unless each bar requires a significant amount of your work (as opossed to scripts extracting the figures from the imatest and producing the input data for the plot program).  Never played with imatest myselft so I don't know that one.

 

By the way, did you ever play with the 40 MP mode of the Oly E-M5ii?  Do you get similar to that Canon when using a top lens of that system?

[dave9t5]

There is nothing in my view, which prevents publishing all four bars - unless each bar requires a significant amount of your work (as opossed to scripts extracting the figures from the imatest and producing the input data for the plot program).  Never played with imatest myselft so I don't know that one.

 

By the way, did you ever play with the 40 MP mode of the Oly E-M5ii?  Do you get similar to that Canon when using a top lens of that system?

 

+1 for including both Center and Near-center (off-center?) along with Corners....and dropping extreme corners.  

 

Always found the Extreme Corners mostly useless data but also misleading...as has been discussed periodically in these forums.  Extreme corners seems to be a legacy of testing FF coverage lenses on low-res APS-C sensors and trying to force data to find some minor differences among lens-sensor combos that all performed similarly.  Consequently, Extreme corner data is over-weighted at PZ.  

 

I've never had a photo where I ever needed any performance out of the extreme corner.  It's such a small area of the photo, but given far too much importance by PZ.  I always ignore the results from it.  

 

Meanwhile, there is no data from the truly important golden-ratio / rule-of-thirds parts of the frame.

 
Center / Off-Center / Corners provide much more useful spread of data across the frame.  Also avoids the perception of such extreme drop-off that you are currently getting with high-res sensors for Center / Corner / Extreme Corners.

 

Thanks for considering my input.

[Brightcolours]

The corners are not that extreme, in PZ tests... Case in point, the EF 35mm f2. Its real (extreme) corners are very soft, very sudden. But the PZ corners are just outside of that, showing relatively good results.

 

I too find not testing the center a bit odd.