LumoLabs: Pentax K-5 low light focus with firmware upgrade 1.03

Fig.1: "Caribbean Sunset".
Why doing measurements is no substitute to taking photos (taken while I type) ;)

The current story is continued from the blog article Pentax K-5 low light focus after Pentax has published the version 1.03 firmware upgrade officially featuring an improvement of focus operation in low light.

When the 1.03 update was released, many users reported positive findings. So, I was optimistic to find that the low light focus issues are a thing of the past. Unfortunately, this isn't what I eventually found.

Please refer to the updated version of the complete paper for my findings:

• www.falklumo.com/lumolabs/articles/k5focus (HTML version)
• www.falklumo.com/lumolabs/articles/k5focus/K5Focus.pdf
  (PDF version for download and offline reading)

Summary:

Basically, I find no significant difference between firmware versions 1.02 and 1.03 when using my testing scenarios. This means that there is no significant improvement if:

• The background is white
• The target has high contrast
• The AF assist light is disabled

Under the above circumstances, the K-5 auto focus starts to lock focus in a false front focus position at tungsten light levels below 0 EV for a fast lens (like f/1.8) or below about 2-4 EV for slower lenses. This did not change at all with release 1.03. Note that the above light level values would read 2-3 steps higher with a target like a Caucasian skin (18% gray).
Moreover, I found that at low enough light levels, the 1.03 firmware seems to rely more than 1.02 on color information outside the direct selected AF spot. With a uniform color distribution, this can lead to somewhat improved results in the vicinity of the focus shift transition. OTOH, the focus system is more easily fooled by features with a singular color. Overall, the advantages may dominate and the effect is small anyway. Version 1.03 still doesn't seem to make use of white balance information, either manually or automatic.

If I believe that Pentax improved the low light focus situation for many users of the 1.03 firmware, then the progress must be bound to any of the following:

• AF assist light engages more actively (not tested by myself).
• Focus improved in the presence of dark backgrounds (not tested by myself).
• Focus improved with low contrast or dim focus features (not tested by myself).

As I don't have a version 1.02 K-5 anymore, I cannot find it out. But any progress brought by the 1.03 firmware upgrade is limited rather than universal.

As much as I applaud Pentax to having addressed the problem, as much I am a bit disappointed they didn't dig deep enough to address the root problem: that the K-5 AF subsystem locks focus even in a situation where the colorimetric sensor(s) fails to determine the light color. It simply shouldn't lock focus at all then. Or ask for user assistance like a priming shot. The light-sensitive AF system of the K-5 has too strong a color dependency to autofocus if the color of the focus feature remains unknown.

A better workaround than 1.03 in firmware is feasible and should stay on Pentax' agenda.

Please, read the full paper linked above if you have further questions and come back here to leave comments or questions. General comments about the issue should still go to the general (earlier) article while firmware release 1.03 comments should go here. Thank You.

LumoLabs: Pentax K-5 low light focus

Note: I updated the article to version 1.3.
2011, March 27.

Note: I updated the article to version 1.1.

I think, the feedback and experiments done after version 1.0 helped to understand the observed focus behaviour rather well and version 1.1 takes this into account.

An important (new) aspect is a description how low key studio photography, esp. when combined with an f/4 lens (or slower), can lead to inaccurate focus.

2011, March 07.


Fig.1: Accuracy of the Pentax K-5 phase detect AF vs. luminosity in EV. The above chart includes all measurements, i.e. various lenses, light colors, distances and apertures. The accuracy is measured as deviation of the focal plane from the sensor plane, in µm.

In modern times, each new release of a digital SLR camera seems to be accompanied by teething troubles. This applies to all makes across the board. Sometimes, they are fixed quickly by the vendor, like the “Green Line Syndrome” video issue in the Pentax K-7, the “Hot Pixel” video issue in the Nikon D7000 or the “String of Pearls” stain issue in the K-5. Sometimes, they aren't like the shutter-induced blur issue with the K-7 which LumoLabs succeeded to document in this blog.

Currently, there are two remaining teething troubles for the K-5 which are widely reported: wrong PTTL exposure with external flashes in some situations (confirmed in writing by 3rd party flash makers, even for Pentax' own flashes). And a systematic wrong lock of autofocus in low (tungsten) light.

Therefore, LumoLabs has decided to have a closer look at the issue. After careful evaluation and many hundred test shots we found the issue to be real. Pentax has unofficially reported to work on the issue. The pressure is on them to address the issue and the author hopes that our findings may contribute to their efforts. Buyers of the K-5 must be able to be confident that the issue is fixed sooner than later.

A preliminary copy of the paper was provided to Pentax earlier this week and the head of Pentax Europe officially receives a printed copy today. I have been told that Pentax engineering will receive a copy too.


The study

The results are too complex to be presented in the scope of a blog. Fig.1 above provides a first idea of our work.

Please refer to the complete paper for our findings:

• www.falklumo.com/lumolabs/articles/k5focus (HTML version)
• www.falklumo.com/lumolabs/articles/k5focus/K5Focus.pdf
  (PDF version for download and offline reading)

You'll gain a deeper understanding of Fig.1 too ;)

In a nutshell, this is what we find:

1. The K-5 as it presently ships indeed has a flaw in its phase detect autofocus module or software which causes it to front focus in low light below a lens-dependent threshold in EV.
2. If it does, it seems to consistently focus ≈ 255 µm behind the sensor plane (although with a significant ± 75 µm scatter of results which is about twice as large as the normal scatter of result).
3. Faster lenses seem to keep working in lower light but of course, are prone to more blur when the front focus does eventually happen. Slower lenses can already start to front focus at light levels metering as 4 EV or 6 EV even. A fast lens may work down to 0 EV in white light.
4. Light sources other than daylight emphasize this problem as they simply appear darker to the AF module. Moreover, it seems to be moderately color blind for red which further emphasizes the effect in deep tungsten light.
5. The effect is real and can negatively impact the daily work of a photographer.  On the other hand, it is possible to run into a low light tungsten situation without the problem.
6. The paper clarifies conditions to hit or avoid the issue. White light (halogen is not white enough though) and a wide lens stopped down help to work around the problem. AF assist light typically doesn't help though. But an LED flash light does. ;)

It may be interesting to note what we did not find: There seems to be no strong dependency on the light color. Except that colored light causes the transition to happen earlier as it means less usable light for the AF module. It seems to have a somewhat low sensitivity esp. for red light. Moreover, there seems to be no dependency on actual aperture or distance.

Move focal plane by 0xFF µm? Yes, do it so! :)

This may describe what's going on behind the curtain: a µm-valued variable becomes 0xFF (255) and causes a false shift of the focus plane by 255 µm. I call it the +0xFFµm hypothesis. :)
- Is it likely? No.
- Is it possible? Yes.

Please, read the full paper linked above if you have further questions and come back here to leave comments or questions. Thank You.

K-7 final firmware and tungsten high iso noise

I have just received the final firmware v1.00 and wanted to make available some samples shot at ISO 400-6400 in tungsten light (2900/2950 K, using Auto WB). Shot with the kit lens and AF focussing onto the SD card in the center.

This is a preliminary article and my evaluation of noise is pending. Also, the subject is not ideal (colors missing, no gray card). I was in a hurry. The ball's surface has a subtle texture looking like noise but not being noise (in iso 400)! So, this may be used to compare denoising artifacts.

The camera was set to "natural" (default is "bright") and default settings otherwise (i.e., sharpness -1, NR medium, NR start iso800).

The images are here (out of camera or Lightroom standard settings):

K-7 Tungsten high iso noise


And the DNG raw files are here:

http://www.file-upload.eu/download-1709906/IMGP0006.DNG.html
http://www.file-upload.eu/download-1709919/IMGP0007.DNG.html
http://www.file-upload.eu/download-1709926/IMGP0008.DNG.html
http://www.file-upload.eu/download-1709937/IMGP0009.DNG.html
http://www.file-upload.eu/download-1709948/IMGP0010.DNG.html



Disclaimer:
Preliminary test, firmware is final but the sensor in my preproduction camera is said by Pentax to be different from the final one!

This is very warm tungsten (below 3000 Kelvin). In daylight, the noise would be much less. Also, ISO6400 is not a fully advertized feature.