[falconeye]

Do I need to whip this URL out again:

Autofocus Torture Test Updated (Canon 1D MKIV & Nikon D3S added) | Pro Photo Home

Thanks for the link.

I am currently searching for a good AF.C testing protocol.

The kind of test you cite seems to be sort of a standard: have an unaccelerated standard subject, camera with tele on tripod in AF.C, center spot and count the # of in-focus images. I know at least 4 such tests, the one you cited, French Chasseur d'Images, German ColorFoto and my blog. And I'm sure there are more.

Let me call this test LS3 test (Linear Subject Subjective Sharpness).

Unfortunately, it doesn't translate into real-world experience. Many pro photographers, e.g., sports, BIF and running dogs (seems to be a market...) report differing results (if I shall believe forums). Typically, Canon ranks top in LS3 while in real life, Nikon seems to be able to deliver the one shot Canon can't. Also in LS3, the K-7 scores very close to the D300 while in real life, it can't compete. BTW, Sony scores good in LS3, typically ahead of Ccanon when tested. I've no idea what the experience is in real life.

The 7D seems to have a very good AF.C. About 2nd within Canon. But it's hard to say how it will compare against D7000 and K-5 in real life.

Therefore, I am at the research of a real-world AF.C testing protocol. Any ideas?

[rparmar]

I can make a comparison! The one thing I have never particularly liked about Nikon is ergonomics. Of course they are a lot of better than Canon but Pentax gets it better!

I agree but Pentax is not yet perfect. The LV button is in the wrong place and in fact should go back to being a position on the On/Off switch so it is not hit by accident. This would allow the back buttons to be operated by feel once again, as was possible on earlier models (eg K20D).

[rparmar]

Neither do I, that's why I refuse to subscribe to the 'bokeH' fad. It's just deliberate blur that only serves to confuse the human eye and brain (which are biologically designed to try and 'translate into clarity', for our own benefit, such as safety). Bokeh is bleah! like drunken focus.

Bokeh is the quality of the out of focus blur area. There is no "bokeh fad" but maybe you mean the rage for low DOF images?

For what it's worth, our eyes see only one small depth of field slice at a time. So actually the camera with large apertures does similar, and those photographers who crave small DOF are simply shooting what the eye sees at one moment. Our brain does something similar to focus stacking to assemble eye-scanned scenes into something that only appears to have greater depth of field.

It's more complicated than that, but I would beware of making essentialist arguments. Like those who argue against digital sound, unaware that our ear is basically digital in operation.

[Fl_Gulfer]

I sure hope the K-5 is allot better than my K-7 and my K100d in high ISO. I went to a Lynyard Skynyard Concert last sunday and took all my gear. I was very dissapointed with the results from the K-7, the wife's K100d did much better on evening photos. I shot on a tripod and she didn't and she still got better shots.

K-7 on a tripod, unedited click for full size shot.

[Chwisch87]

Well the k-7 should be far far better than the k100d is just about any situatiuon.. Have you ruled out user error?

[falconeye]

For what it's worth, our eyes see only one small depth of field slice at a time. So actually the camera with large apertures does similar

Not true.

Human eye's focal length is ~17mm,
aperture (pupil surrounded by iris) varies between 3 and 9mm depending on the individuum and available light.

In full light, it's 3mm (4mm max.). So, the human eye's f-stop at day time is about f/5.6. When young, it can reach f/1.8 in a dark night though.

Taking half an Airy disk at f/5.6 as CoC (4µm) we have a hyperfocal distance at 17mm such that everything from 6.4m to infinity is in focus.

At armlength (80cm), the DoF still is 10cm. Not exactly "a small slice".

[melander]

I sure hope the K-5 is allot better than my K-7 and my K100d in high ISO. I went to a Lynyard Skynyard Concert last sunday and took all my gear. I was very dissapointed with the results from the K-7, the wife's K100d did much better on evening photos. I shot on a tripod and she didn't and she still got better shots.

K-7 on a tripod, unedited click for full size shot.

Dont blame that shot on the camera, I would look behind the camera instead

[kittykat46]

Not true.

Human eye's focal length is ~17mm,
aperture (pupil surrounded by iris) varies between 3 and 9mm depending on the individuum and available light.

In full light, it's 3mm (4mm max.). So, the human eye's f-stop at day time is about f/5.6. When young, it can reach f/1.8 in a dark night though.

Taking half an Airy disk at f/5.6 as CoC (4µm) we have a hyperfocal distance at 17mm such that everything from 6.4m to infinity is in focus.

At armlength (80cm), the DoF still is 10cm. Not exactly "a small slice".

The highly selective detail vision of what the human eye perceives is not related to focal length or DOF. Its mainly due to Biology.

The area of the human retina called the fovea, which has acute resolving power, both for colour, detail and dynamic range, is less than 1mm in diameter. The rest of the retina moving away from the fovea has diminishing detail perception.
Since we can rotate both our eyeball and the head, we can "see" various parts of the scene in great detail.
Its like a sensor which has incredible resolution and dynamic range in one tiny spot, and the rest of the sensor has pretty low resolution.

[falconeye]

The highly selective detail vision of what the human eye perceives is not related to focal length or DOF. Its mainly due to Biology.
The area of the human retina called the fovea, [...]

What makes you think I didn't take this into account?

This is why I took half the Airy disk as resolution parameter. Biology doesn't break the laws of nature.

To make you feel better ... the angular fovea resolving power usually quoted (1.7 arcminutes per linepair which is a 120% vision) translates to a 4.2µm feature size (8µm line pair) which is actually worse than my 4.0µm diffraction-based limit I used in my previous post.

On a side note, the human eye does not normally reach the diffraction limit as it has significant spherical abberation. At 3mm, it is minimized and the eye reaches its best resolution. With its sweet spot at f/5.6, it isn't unlike good prime lenses


The fovea just makes best possible use of the lens' center performance. Better vision would need more lens elements (not yet invented by mother nature, or the big creationist planner is an idiot) or a larger sensor.

The human retina has a surface of ~650+ mm^2 (1.1 crop factor of 35mm full frame) which is why we can consider it a full frame camera.

So, a better vision would require the biological equivalent of medium format or, like the eagle, a second fovea optimized for tele view. There, the eagle resolves the equivalent of 1.7µm features (source found) or better (no sources found).

But larger cameras exist too. E.g., the Colossal Squid has eyes with 27cm diameter as opposed to our 22mm and 150x the surface which should account for 7 stops better high ISO performance

[Eruditass]

Thanks for the link.

I am currently searching for a good AF.C testing protocol.

The kind of test you cite seems to be sort of a standard: have an unaccelerated standard subject, camera with tele on tripod in AF.C, center spot and count the # of in-focus images. I know at least 4 such tests, the one you cited, French Chasseur d'Images, German ColorFoto and my blog. And I'm sure there are more.

Let me call this test LS3 test (Linear Subject Subjective Sharpness).

Unfortunately, it doesn't translate into real-world experience. Many pro photographers, e.g., sports, BIF and running dogs (seems to be a market...) report differing results (if I shall believe forums). Typically, Canon ranks top in LS3 while in real life, Nikon seems to be able to deliver the one shot Canon can't. Also in LS3, the K-7 scores very close to the D300 while in real life, it can't compete. BTW, Sony scores good in LS3, typically ahead of Ccanon when tested. I've no idea what the experience is in real life.

The 7D seems to have a very good AF.C. About 2nd within Canon. But it's hard to say how it will compare against D7000 and K-5 in real life.

Therefore, I am at the research of a real-world AF.C testing protocol. Any ideas?

Very interesting that the K-7 does nearly as well as a D300 in a LS3 test. I'd be interested in reading some of those tests.

One note about that link. They used default settings, where the Canon defaults to focus priority, and the Nikon defaults to shutter release priority. You can see all the angry Nikon people in the comments

[Arpe]

Here let me tell you about shooting pictures at settings i never use ... oh i could but ... i never shoot pictures at anything higher than 1600 ... Nor does anyone else for that matter.

Why is it that people say, to paraphrase, "I don't do it this way, so nobody does". Dumb. I have hundreds of photos at ISO3200 and over.

Well the point of my "nor does anyone else for that matter" is basically the fact of the matter is, if you are looking to take your best photo possible, you are going to stick to around ISO 200 to 320... no matter what body you use. Your best photos are taken under ideal light conditions. Most of us are not looking to take a picture at even 1600 and expecting to get into the PPG. And even for indoor photography, IMO, its always best to just bring out a nice strobe.

Strobe? Yeah I'm sure the players would love that!! Again, just you can, doesn't mean everyone can.


I recall dpreview the raw noise graphs for the K-7, D300 and 7D only really separated above about ISO1600. It seems silly comparing out of camera jpegs as the settings are so adjustable in camera.

[falconeye]

Very interesting that the K-7 does nearly as well as a D300 in a LS3 test. I'd be interested in reading some of those tests.

That was in a standardized 100 km/h car tracking test which I reproduced last year for the K20D and K-7 in my blog. A K20D and D300 were compared directly in the source I took the test procedure from.