kye Posted November 3, 2022 Share Posted November 3, 2022 1 hour ago, Django said: AF performance can absolutely be quantified. How it levels to your particular needs is subjective though obviously. Generally speaking subpar AF performance is simply an AF system that hunts/pumps or/and doesn't stick to your subject. Then you've got second gen PDAF that detects left/right eyes of various subjects. Object, animal etc detection. And finally tracking. Canon/Sony have tap to track which Fuji is still lacking. I firmly believe that almost everything can be quantified, but this one falls very far beyond the point of what is practical. The GH5 tests I saw with the person stepping into and out of frame had the AF recognise a face and change focus across a range of reaction times - sometimes it was fast and other times reluctant, and occasionally the person would just stand there being ignored like a camera nerd at a high-school dance. There aren't any easy way to quantify this. GH5 testers couldn't even get the test to replicate, providing a number of hilarious examples were the person was walking around saying how bad it was and it tracking them just fine, and other testers saying it was really good and it doing quite badly. One YouTuber who got the C70 when it first came out admitted in a follow-up review that he had to stop using the C70 until it had a firmware update or two because when it first arrived it had trouble recognising faces of darker-skinned people. IIRC he had to hire something to use on commercial shoots because the C70 wasn't ready yet. Canon can't even test their AF properly and it's one of their key brand differentiators! If you were to quantify AF performance, not only would you have to have a dozen or so metrics (speed to recognise a face, how out-of-focus the face can be before it recognises a face, maximum tracking speed, how much of the face has to be visible, how far around the side of the face it detects, how bad the lighting has to be for it to recognise a face, etc) but you'd really struggle to quantify the GH5-style lack of reliability except to have an enormous sample size. Peter McKinnon made a promo video for his new product, and at the 6:12 mark, the camera goes from focusing on the object: to focusing on his face: the two frames above are 2 frames apart. Why did the mighty Canon PDAF randomly choose that moment to change focus to his face from the largest object in the centre of the frame? Heck knows, but there were even previous frames where more of his face was showing and it didn't choose those times to change focus.... Here's the video linked to that time - judge it for yourself... That's an AF problem right there. People tend to think that Canon PDAF or Sony eye-detect PDAF are perfect but in reality they stuff up from time to time, and they tend to think that GH5 is completely useless when it actually gets things right quite a large percentage of the time. I've seen shots in vlogs where Canon PDAF cameras just randomly focus on the background when the persons face was visible the whole time. They're rare, but I've seen at least two that made it to the final edit - we can only guess how many others ended up being cut. Any methodology that quantifies AF performance would be useless if it ignored the lack-of-reliability problem (because it would declare the GH5 AF to be great when it's obviously not) and it would be wrong if it gave Canon and Sony a perfect score when they obviously aren't quite there, despite being impressively close. Sure, you can quantify some aspects of AF performance, but to be even remotely useful, you'd have to test so many variables and some of them would require such incredibly large sample sizes that it just wouldn't be practical. PannySVHS and IronFilm 1 1 Quote Link to comment Share on other sites More sharing options...
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.