DxO: F-stop blues

[bakubo]

I read this article on the dxomark site and thought it was interesting. Maybe others will too.

For years, lens makers have fought hard to market lenses of wider and wider aperture. Wide apertures (e.g., f /1.4 instead of f /2) but a series of measurement published on dxomark.com cast some doubts on the real benefits, for digitally equipped photographers, of these progresses.

http://www.dxomark.com/index.php/Our-pu ... stop-blues

[David Kilpatrick]

That's fairly obvious, but would not apply to sensors without microlenses. Microlenses resemble an Acute Matte focusing screen, so will show an increasing loss of effective aperture at anything wide than a certain setting such as f/2.8 or f/4. They will also show this more so at the edges (vignetting).

What I find surprising is that the A350 (an odd camera to choose for this test) is claimed to have a 1.4 stop additional gain applied when using an f/1.4 lens, the highest figure of all the camera shown. That would make its 3200 a genuine ISO 8000 if you use a 50mm f/1.4.

As ever with DxO Mark, the information is fascinating but incomplete as it does not test different focal lengths (an 85mm f/1.4 should be better than a 35mm f/1.4) or different telecentricity (Olympus f/2 or f/1.4 lenses may for example NOT suffer the same way but they don't bother even to test this).

And they fail to comment that the use of fast lenses for differential focus effects is unaffected, so it is still worth having and using f/1.4 lenses if you want f/1.4 depth of field.

David

[agorabasta]

Those DxO guys are like little children opening the world of wonders anew for themselves.
So now once again they present those results as a real revelation, while everybody else knew it well before the DxO was even born. And once again they do it in a completely unscientific way. Why didn't they try using the same lens for all the bodies in this test? Why didn't they then even try to analyse the effect of telecentricity which is the most important lens design property affecting the silicon sensor transmission loss at oblique angles of incidence?

And most importantly - why calling it "F-stop blues" when it's clearly not the lenses themselves but the sensors causing that transmission loss?

[bakubo]

What I find surprising is that the A350 (an odd camera to choose for this test) is claimed to have a 1.4 stop additional gain applied when using an f/1.4 lens, the highest figure of all the camera shown. That would make its 3200 a genuine ISO 8000 if you use a 50mm f/1.4.

Yes, strange, but no more strange than using the very old Nikon D200. I suspect it was a matter of what cameras they happened to have around at the time.

And they fail to comment that the use of fast lenses for differential focus effects is unaffected, so it is still worth having and using f/1.4 lenses if you want f/1.4 depth of field.

My guess is that they assume a certain amount of knowledge of the reader so didn't explain the difference in dof.

[agorabasta]

They put the figure 1.4 at the end of a350 plot as the f-number and not as loss/amplification value. The extra gain there is slightly above 0.5 stop as shown in that plot.

Once again shows that they cannot even make their plots look consistent...

[bakubo]

agorabasta, it would be great if you would do this properly and post the results here.

[Greg Beetham]

I got the impression it was all theoretical, not actually plugging various cameras into a test bed and getting real value readouts.
Greg

[agorabasta]

It all is not theoretical, it's quite practical and well known like forever. And it's not the wide aperture that causes the effect. It's simply that oblique rays are not well registered by the reflective silicon surface. The microlenses help. Other sensor toppings mostly do harm.
The right way to test the cams for that property would be to have a thin collimated white beam hit the sensor at the most oblique possible angle and then compare the readout to the straight-on direction of the beam readout at the same shutter speed and ISO. The test obviously doesn't need a lens at all.

And btw, that property of the silicon sensors has one nice side effect. It suppresses the brightness of the image of aperture edge thus making the blur look softer, sorta a little STF on the cheap.