SCoombs
Offline
Upload & Sell: Off
|
 Re: Z8 extremely unreliable/inconsistent AF in mildly low light | |
I have an update with some data here which I expect some users to no doubt treat as an "I told you so" moment, but I'd urge people to read the following carefully before jumping to that conclusion.
I went ahead and painstakingly (and it truly was really incredibly boring) go through every one of my 1000+ shots from Friday night and to make a judgment as to their focus and record the focus mode and quality of lighting for each. A few methodological notes:
First, I only rated photos of people.
Second, I skipped a couple dozen photos where, either due to noise or other issues, I did not think I could make a meaningful judgment as to whether a photo was in focus or not.
Third, I skipped photos which (for any focus mode) had the subjects in focus but the focus point off in space somewhere so that I could not determine whether or not the shot was actually in focus or if it turned out in focus because the system actually missed the erroneous thing the point was on.
Fourth, I skipped a whole segment of photos where people were running around in very dark conditions because I got a 0% hit rate using all modes during this period. It didn't even signal focus and simply miss as in all other cases, but in this case it just very slowly hunted and never even acquired any sort of focus.
I divided the photos into two lighting situations. First there was a more well lit "cocktail hour" room, and second was the main event room which was darker. In total I rated 440 photos.
Results are as follows:
In the brighter room, Single point was used 42 times and gave in focus results 83% of the time.
In the brighter room, dynamic area was used 46 times and gave in focus results 89% of the time.
In the brighter room, subject detection (as Auto-area AF) was used 114 times and was in focus 73%.
In the darker room, single point was never used.
In the darker room, Dynamic was used 126 times and was in focus 82% of the time.
In the darker room, subject tracking was used 112 times and was in focus 65% of the time.
Overall, Single point hit 83%, dynamic hit 84%, and subject tracking hit 69%.
Your first thought may be that these are much greater hit rates than the 30% I reported. That is true, but the reason for this may I think be found in snapsy's tests. When testing at home when I was getting 30% hit rates I was almost always photographing people around or less than 2 meters away. Snapsy seems to have found that this situation may be the worst for the system's performance. At the event, people were almost always more than 2 or even 3 meters away. Additionally, snapsy seems to have found that photographing people head on is less likely to provoke the issue as compared to shooting people at angles. In the event, not as many of my subjects were at angles.
Your second thought may be that subject tracking actually performed somewhat well compared to the other modes, but I'd add a very important caveat to that: while I didn't record it in a formal way, the in focus shots from dynamic area were much, much more often what I'd call critically sharp. The subject detection focus gave me a very large number of shots that I'd call "soft in focus," where they were probably close enough to call in focus but were still below the standards people are generally going to expect from a modern camera. There were a few subject tracking shots that hit very nicely but for the most part they were pretty soft/borderline, whereas the majority of the dynamic area in focus shots were just spot on.
Another way of sayin this is that most of the subject detection hits look like they're in focus when viewed on their own but if you put them side by side with the dynamic area shots you might be tempted to say say they missed.
I didn't track this formally because I didn't really start to consciously recognize the trend until I was about halfway done, but I'd estimate that if I went back and evaluated for something like "tack sharp" only I think the subject detect would sit at around 20% -30% and dynamic area would sit at around 65-70%.
So to summarize my current findings:
* Dynamic area seems pretty reliable in low light, though not nearly as reliable as the "my hit rate is almost perfect" that I see some saying
* Subject detection seems to be roughly a 30% miss rate if we have low standards for what constitutes in focus, closer to a coin flip or worse if we want critically sharp photos
* All of this seems to apply when shooting subjects at distances of 2-2.5 meters or greater
* When shooting subjects at distances of 2-2.5 meters or less, the hit rate for subject detection appears to drop dramatically.
Finally, one interesting note: in the mix of all of these were two very, very sharp photos taken in auto area AF where the system had been unable to detect a face (though it was very clear) and resorted to just focusing with the matrix of focus points. These may have been the sharpest photos I got out of that mode all night.
|
