X-H2S subject tracking miss

Truman Prevatt wrote:

michaeladawson wrote:

I'm not surprised to hear someone report having problems like this. I'm not saying the X-H2S has an issue. I'm not saying that this is user error.

What I am saying is that you have a relatively low detailed subject, a gray body and head at too long a distance, sitting in a sea of very detailed and high contrast grass.

I would expect an advanced camera like the X-H2S to be able to easily nail this shot. But I'm also not surprised in the least if someone reports that it's not working.

The Nikon Z9 also has many reports of scenes where the camera's AF system will fail to grab focus on what to the photographer looks like an obvious subject and instead locks onto the background.

Nikon has been making continual tweaks to the advanced AF system of the Z9. I'm sure we can expect to see many refinements to the X-H2S AF system.

No camera knows that the subject of interest is a bird. If it is set to bird, all it knows is a set of contrast values to input to the algorithm to compare with the subject. This template is not a bird - it is a set of numbers and there are an almost infinite number of ways the camera can find matches to the template with statistical certainty.

Subject detection is a statistical process. AF is a statistical process on top of that. Tracking is yet another statistical process on top of those two. Every statistical decision has the opportunity to make two types of errors. One cannot simultaneously minimize both errors, the first, Type I - not detect followed by A/F followed by track on a subject that is there and the second Type II - detect followed by A/F followed by track at bird in this case that is not there or the grass behind the bird.

All the "deep learning AI" will provide is a different way to calculate the templates that is hopefully better. On the other hand the Type I and Type II errors are omnipresent.

The real world presents a nasty environment for this type of process to work. A bird against a clear sky is pretty simple. A dark bird sitting on lush tree grass, or a dark bird in a heavy tree cover is going to tax most cameras.

The trick is to guide the camera to focus on the subject and then the subject detection is much more likely to detect the subject and/or it's eye and stick to it.  This means one must use a focus method that starts by focusing on the subject such as zone or single point.  Using wide area requires the camera to guess as Truman points out.  Hoping that the camera will find the subject will sometimes fail as home is not a method.

Morris