AF review of Eos 1Dx Mk3 by Fro

[Jack Douglas]

And price

I hope so but this may be the highest priced 5D ever.

Jack

 

[koenkooi]

I hope so but this may be the highest priced 5D ever.

With or without adjusting for inflation?

 

[Kit.]

I don't see how you could differentiate between a helmeted head sitting in an open cockpit and a basketball.

By the fact of the presence of the cockpit.

That involves gathering and labeling some amount of "ground truths" for the neural net to learn, yes.

 

[privatebydesign]

By the fact of the presence of the cockpit.

That involves gathering and labeling some amount of "ground truths" for the neural net to learn, yes.

Thats too literal, what about a helmeted head on a luge, or walking across a pit area, on a motorbike, slalom skiing etc etc....

 

[koenkooi]

I love how the camera momentarily locks onto the basketball a couple of times. I wonder if it's seeing a face in there or if the deep learning neural net has learned to recognize balls as pertinent objects in their own category.

It also likes that pom-pom a lot, maybe it likes contrasty spheroids?

 

[pj1974]

It also likes that pom-pom a lot, maybe it likes contrasty spheroids?

I am thinking that the intelligent AF actually locks onto an object, which is a combination of being a moving spheroid and a contrasty spheroid.

The reason I write this is because in a number of Jared's/Fro's photos, the blue AF square jumped from being locked onto the stationary person's face (which was still facing the camera enough to be detected as a face) to locking onto the ball or pom-pom. The ball or pom-pom had then become the fastest moving object in the composition, and also was reasonably contrast against the respective backgrounds.

My $0.02.

PJ

 

[koenkooi]

I am thinking that the intelligent AF actually locks onto an object, which is a combination of being a moving spheroid and a contrasty spheroid.

The reason I write this is because in a number of Jared's/Fro's photos, the blue AF square jumped from being locked onto the stationary person's face (which was still facing the camera enough to be detected as a face) to locking onto the ball or pom-pom. The ball or pom-pom had then become the fastest moving object in the composition, and also was reasonably contrast against the respective backgrounds.

My $0.02.

PJ

This is why I dread changing AF options like tracking and responsiveness, I'm too lazy to setup repeatable tests and take enough time to keep rereading the manual.
Maybe Jared/Gordon/Matt will do a youtube video on it, would save me a lot of time

 

[Kit.]

Thats too literal, what about a helmeted head on a luge, or walking across a pit area, on a motorbike, slalom skiing etc etc....

The same. The more scenarios you are providing for the net to recognize persons and find their head positions during training, the better. We are talking about no less than tens of thousands of pictures anyway.

As a bonus, you might be able to estimate where to focus to keep the head in focus even if the head is completely obscured at the moment.

 

[Viggo]

Why aren’t we talking about the fact that I can’t see any blackout or lag even with the mechanical shutter here?

and holy smokes that blue square moves on that pompom! That is insanely fast AF tracking.

 

[koenkooi]

Why aren’t we talking about the fact that I can’t see any blackout or lag even with the mechanical shutter here?

and holy smokes that blue square moves on that pompom! That is insanely fast AF tracking.

I really, really, really hope all new R models will get similar performance.

 

[Bennymiata]

I think Fro was very impressed and I don't think he was expecting it to be so good.

 

[Jethro]

I really, really, really hope all new R models will get similar performance.

The 1DX III gets an extra DIGIC dedicated to AF, doesn't it?

 

[koenkooi]

The 1DX III gets an extra DIGIC dedicated to AF, doesn't it?

The wording makes it seem that extra DIGIC only processes the data from the external AF sensor, so it wouldn't be needed on mirrorless.