I don’t think Generative AI will have anything to do with it unless it’s on-camera post processing.I'm not sure how generative AI would work in this situation. GenAI takes your inputs to then iteratively improve the focus selection results for next time. It isn't clear to me how the individual camera would identify which images are significantly better than others to refine AI autofocus for the next shot. Normally we do that in post to pick the "best" image and delete ones where the focus is off however it was determined by the original algorithm. Maybe using the Rate button but I am not sure how many users use this feature.
I had the situation on the weekend shooting indoor karate training session with pairs of students doing sparring drills together with spectators sitting down a meter from the mats.
I often needed to select the correct face using the Set button to get the right one at the center point to start tracking. Note that joystick left-right selection wouldn't have worked due to the sheer number of heads potentially to select.
The tracking was great once it locked onto the right subject.
I can't see how genAI would assist in this scenario ie all the heads were options so one wouldn't be better than another.
That said, if I could train it to detect the rhinopores of nudibranchs then that would be very useful
I think they’ll use some AI-ish machine learning to develop subject and scene recognition, and then select and refine algorithm parameters accordingly.
“Oh, this looks like an animal approaching the camera. Select AF points where its nearest eye likely is, and prioritize for closing distance.”
Use of “AI” would be on the development side, not the camera side, in my estimation. They certainly wouldn’t use Generative AI to train the system; that would contaminate the dataset (“make a picture of a tiger and then learn to recognize a tiger using that picture” is bad juju). Conceivably they could use Generative AI to write code (“track the tiger’s position,”) but canon isn’t exactly known for its software prowess so it isn’t likely they have a model capable of that (vice OpenAI or Google).
Last edited:
Upvote
0