In this patent application (2024-073200) Canon is looking at improving the racking of the lens from near to far to find focus quicker.

As you may be aware, major defocus happens when the contrast is too far out of focus that it's just soft with no visible contrast. When this happens, it's nearly impossible for the camera to judge how far to move the lens to achieve focus. This is when the lens goes “hunting” for it, which usually is a dreadfully slow process as it racks in and out from near to far.

Canon in this patent application, looks to change that with some intelligence in the camera to set defocus ranges and use that to quickly isolate where the plane of focus is.

Major defocus is a major problem with mirrorless, so Canon finding a resolution to one of the greatest weaknesses of mirrorless would be a wonderful thing for us users.

As with all Patent Applications, we may or may not see this implemented but it's encouraging that Canon is still researching ways to improve this and not just going “Meh, it is what it is.”

Source: Japan Patent Application 2024-073200

Go to discussion...

5 comments

  1. Please also implement more AI in object recognition when focusing. Would be great to be able to recognize birds among branches and focus on that instead of nearest twig.
    • 0
  2. Please also implement more AI in object recognition when focusing. Would be great to be able to recognize birds among branches and focus on that instead of nearest twig.
    That would be a big change compared to how it currently works, detection and focus are not coupled and the focus is open-loop. This is how I think it works, based on first hard observation and people complaining on the interwebs:
    1. The camera looks at the scene, detects a bird, draws a box around it
    2. An eye! Camera draws a tiny box around the eye, which sadly also includes a twig behind the bird.
    3. The camera tells the focus system to focus on the area inside that tiny box
    4. The focus system says "ah, DPAF estimates that's 2.12 meters" and tells the lens to focus to 2.12m
    5. The lens focuses to 2.12 meters, or the closest step its motors and encoders can do
    6. The lens tells the body "I'm at the distance you told me"
    7. The body starts the exposure
    The issue that you mention, bird vs twig, is due to the detection and focus system being separated between step 3 and step 4. DPAF picks the twig in step 4. And the RF70-200 f/2.8L has shown that step 5 depends on the calibration of the lens and that might need a firmware update when it consistently misfocuses by the same amount.

    One way of alleviating this, is to have the body do an extra detection between step 6 and 7, which uses contrast (aka pre-DPAF on-sensor focus) to judge critical focus. My M6II has a menu option for "Release priority" vs "Focus priority", which would be a good way to toggle this behaviour. This would also help @neuroanatomist to focus on perfectly straight features in the 'wrong' orientation.

    It would be even better if users could do an AFMA style calibration themselves to avoid needing a firmware update to make the lens focus properly, like the RF70-200L needed.
    • 0

Leave a comment

Please log in to your forum account to comment