• site home
  • blog home
  • galleries
  • contact
  • underwater
  • the bleeding edge

the last word

Photography meets digital computer technology. Photography wins -- most of the time.

You are here: Home / a9 / Stacked sensors and autofocus

Stacked sensors and autofocus

March 26, 2019 JimK 20 Comments

Almost all my technical posts are based on research and testing. This one is different: it is entirely speculation.

Yesterday I upgraded the firmware on my Sony a9 to version 5.0, which had just been made available for download by Sony. It is much more than any other Sony firmware release that I’ve ever seen. Sony usually just fixes bugs in their FW releases, occasionally introducing a new feature like their malicious-obedience uncompressed raw support. In that regard, they’ve been like Nikon, and not at all like Fuji, who routinely roll out new features via firmware upgrades.

FW 5.0 for the a9 in invasive. It wipes out all your settings, including date and time. There’s a reason for that; the menu structure is heavily reworked, and there are small improvements and new things spread throughout the new menu system. It still feels like a Sony, which in my mind is not a good thing, but it still getting better. The custome key assignment now works like that on the GFX, and that’s a big step forward. Focus area limits can be set by focus mode, with exquisite granularity.

But the big change is in autofocus. There are new modes, including one that does a great job of tracking moving subjects, and an Eye-AF implementation that is far less cumbersome than the previous one that required assigning and then holding down a button other than the shutter release. You can pick whether the Eye-AF focuses on the right or left eye, or leave it in automatic, which was the only choice before. It’s going to take me a while to understand all the improvements, but playing around with the new autofocus makes me think that Sony has taken a camera that was already no slouch in the AF department to an entirely different level.

If you have an a6400, you’ve got a lot — maybe all, for all I know — of the new AF trickery, but if you have an a7III or a7RIII, for now at least, you’re out of luck. I’m thinking that part of the reason may be that it’s hard on those cameras to read the sensor and the PDAF pixels fast enough. I know the a6400 doesn’t have a stacked sensor, so that runs counter to my guess, but I don’t know how well the a6400 AF tracking works compared to the a9’s new abilities. To make a camera that identifies and focuses continuously and accurately on particular things in the scene — like eyes or like a moving subject — you need to to read the sensor with high refresh rate and low latency. That’s something that the a9 does spectacularly — maybe uniquely — well, even before the FW upgrade.

Sony says that the new FW uses “AI-based object recognition”. I don’t know exactly what that means. Does it mean that it uses neural networks? Or does it mean that the algorithm was derived through training on real-world images, and that, like so many such programs are opaque in their machinations, even to their developers? Whatever I means, I’m intuiting that, like so many similar programs, the more data it gets, the better it will work, so the camera’s ability to read scene luminance (through the normal pixels) and scene depth (kind of, through the PDAF pixels) densely and rapidly will be a boon.

This puts new emphasis on stacked sensors. Sure, they’re great for high frame rates. We’ve seen in the a9 that that’s necessary for good AF, since with a MILC the AF system is blind while the actual photographs is being read off the sensor, and the AF system needs to make up for that by reading the sensor very quickly when it does get a look at the scene. But reading the scene densely enough to make area-based PDAF work and reading it well enough to recognize and track objects — even if they are momentarily obscured — is quite another thing. I’m guessing that this kind of subject recognition will become more prevalent in the future, and that it will be one more spur in the sides of the sensor manufacturers to develop more stacked sensors.

Which brings me to a question: the a9 first shipped almost two years ago. In all that time we’ve not seen another full-frame stacked sensor, right? How come? Is it because the a9 sensor is more expensive to make than originally thought? That the yields are so low that Sony doesn’t have the stomach to do more of those kinds of sensors? That those sensors are not getting rapidly cheaper to build over time? Or that they just don’t see the benefits?

a9

← Lap times as lab testing Speculating on the GFX 100S shadow noise performance →

Comments

  1. roglic says

    March 26, 2019 at 11:50 pm

    Sony say stacked sensor is very expensive + ISO 100 DR of the A9 is bad. I would not buy “A7R IV” camera with stacked sensor and bad DR.

    Reason why A7 cameras dont get the upgrades is either 1) A7 hardware is not good enough or 2) “Canon” strategy of giving only the more expensive camera new features. A6400 proves no need for stacked sensor for new AF. So is number 2 reason.

    Reply
    • JimK says

      March 27, 2019 at 6:33 am

      In my a9, the base-ISO DR is plenty for those things that the camera is especially suited for.

      Reply
    • MIck says

      April 3, 2019 at 1:54 pm

      I own both the A7RIII and the A9.

      DR is better at base ISO for the A7RIII but the A9 is still quite good. And it is better at higher ISO than the A7RIII.

      The difference at ISO 200 is just 0.5 EV, hardly visbible and at ISO 800 they are even.

      That difference is totally acceptable given the performance of the A9 and the targeted user base !!

      Reply
  2. Kevin says

    March 27, 2019 at 1:02 am

    Regarding your question on “AI” for tracking AF, my thoughts are that they are using a new recognition algorithm trained using neural networks or something similar on objects, but this training has all occurred beforehand on Sony’s computers. The final result is incorporated into the camera, essentially serving as a reference to recognize moving people and objects. No training is occurring on the camera, which most certainly lacks the dedicated processing capability.

    As for the lack of other large sized stacked sensors (FF, APS-C), I think you are right about the high costs. Although there must be some margin because I’ve seen the A9 discounted very heavily, even at $3500 (and probably lower), indicating it may be even possible to sell a $3k camera with the stacked sensor. And looks like they are able to make the stacked sensors in large amounts at the 1″ scale for the RX100 IV and up.

    A non-stacked sensor like the one in the A7III might just be so much cheaper for Sony to make that the economics make the decision a no-brainer.

    Reply
    • MIck says

      April 3, 2019 at 1:56 pm

      I got my A9 for what amounts to approx $3300 (Ebay saller with good reputation) and for that money it is a fantastic camera 🙂

      Reply
  3. N/A says

    March 27, 2019 at 5:00 pm

    > In all that time we’ve not seen another full-frame stacked sensor, right?

    and Sony Imaging got exactly one FF new sensor since that from Sony Semi – that is a new 24 mp FF for the new cheapest FF line… so why be surpized at all ? we are yet to see new sensors destined for more expensive cameras where extra cost is better recouped: A7R-4, A7S-3, A9-2 … good chance that A7S-3 will be stacked.

    Reply
    • JimK says

      March 27, 2019 at 5:07 pm

      I didn’t mean just in Sony cameras…

      Reply
  4. Lee says

    March 27, 2019 at 7:47 pm

    Some say a6400 have same processor as A9,but A7III and A7RIII only got a less more powerful one.So they can’t doing those tacking-AF thing.

    Reply
    • JimK says

      March 27, 2019 at 7:48 pm

      References?

      Reply
      • lee says

        March 28, 2019 at 6:08 pm

        https://www.youtube.com/watch?v=GPBz18Z8XTY

        at about 25:15

        Reply
        • JimK says

          March 28, 2019 at 9:00 pm

          Thanks.

          Reply
        • eggry says

          March 29, 2019 at 11:01 am

          the Video is not saying a7R III and a7 III have less capable processor, just saying a6400 brought same processor of a9. I think all four cameras using same BIONZ X and new front-end LSI as far as their marketing materials say. based on knowns, the only obstacle prevents a7R III and a7 III to get tracking AF is just marketing or cannibalizing issue.

          Reply
          • JimK says

            March 29, 2019 at 12:22 pm

            the Video is not saying a7R III and a7 III have less capable processor, just saying a6400 brought same processor of a9

            I noticed that as well.

            Reply
  5. eggry says

    March 29, 2019 at 10:51 am

    I’m curious about RAW compression behave after the update. a7R III and a7 III support uncompressed RAW continuous shooting out of box but a9 has been suffered from compression when shoot continuous even if set as uncompressed while suffering from lower frame-rate too which I think unfair! lol

    Now a9 supports RAW+Extra fine quality JPEG which lacked at a9 before but supported at a7R III and a7 III like uncompressed RAW continuous shooting. Perhaps now a9 output full 14bit, uncompressed RAW in continuous shooting? How do you think Jim? I don’t have tool or knowledge to examine it unfortunately 🙁

    Reply
    • JimK says

      March 29, 2019 at 12:23 pm

      a9 has been suffered from compression when shoot continuous even if set as uncompressed

      I’ve not seen that.

      Reply
      • tree says

        April 25, 2019 at 11:47 am

        I think what eggry meant is that A9 drops to 12bit in continuous mode. Don’t know if this is improved.

        Reply
  6. Eric Calabros says

    March 31, 2019 at 10:38 pm

    Maybe with speed benefits new ADC designs offer they see stacking doesn’t worth the effort.

    Reply
  7. David says

    April 1, 2019 at 3:43 am

    Hi Jim,

    Several comments. A7 III and A7r III will recive an update of the firmware to have the same level of AF characteristics as the a6400: https://www.dpreview.com/news/5057389015/sony-promises-firmware-updates-for-a7-iii-a7r-iii-and-a9-improving-af . But for what I read in the press release, this will not have the real-tracking ability of the a9, for that you will need the stacked sensor.

    Image resource has an interview with a Sony CEO: https://www.imaging-resource.com/news/2019/03/22/sony-qa-cpplus-2019-future-of-autofocus-plus-expect-more-aps-c-bodies , they comment several things:

    1- Yes, stacked sensor is expensive to make… although I asume it is the same process as the mobile phones one, just bigger, I suspect the process will get cheaper in time like CMOS technology.
    2- The AI tracking works uploading a dictionary, that I asume that t is the machine learning result of trainning the algorithm off-line. In the DPreview news they talk that by summer a9 will get a firmware update for tracking animal eyes, that implies the user needs to upload a different dictionary, that I assume that it is for the training process with animal eye samples.

    I have it clear, future Sony cameras will have stacked sensors, maybe not all, like the r series, that thinks in better DR and resolution, but the non-r series… or the s series for video, this can be quite interesting.

    Regards,

    David

    Reply
  8. tree says

    April 25, 2019 at 11:58 am

    Hi Jim!

    Sony mentioned improved color reproduction in the FW5.0 introduction.
    “The camera will more accurately account for subtle changes in light for smoother, more natural tonal gradations in elements like the sky.”

    Did you do any test on this?
    Does it only affect SOOC JPG?

    Reply
    • JimK says

      April 25, 2019 at 12:28 pm

      Did you do any test on this?

      No.

      Does it only affect SOOC JPG?

      I don’t know. Sorry. I never pay any attention to camera JPEGs that aren’t embedded in raw files.

      Reply

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

May 2025
S M T W T F S
 123
45678910
11121314151617
18192021222324
25262728293031
« Apr    

Articles

  • About
    • Patents and papers about color
    • Who am I?
  • How to…
    • Backing up photographic images
    • How to change email providers
    • How to shoot slanted edge images for me
  • Lens screening testing
    • Equipment and Software
    • Examples
      • Bad and OK 200-600 at 600
      • Excellent 180-400 zoom
      • Fair 14-30mm zoom
      • Good 100-200 mm MF zoom
      • Good 100-400 zoom
      • Good 100mm lens on P1 P45+
      • Good 120mm MF lens
      • Good 18mm FF lens
      • Good 24-105 mm FF lens
      • Good 24-70 FF zoom
      • Good 35 mm FF lens
      • Good 35-70 MF lens
      • Good 60 mm lens on IQ3-100
      • Good 63 mm MF lens
      • Good 65 mm FF lens
      • Good 85 mm FF lens
      • Good and bad 25mm FF lenses
      • Good zoom at 24 mm
      • Marginal 18mm lens
      • Marginal 35mm FF lens
      • Mildly problematic 55 mm FF lens
      • OK 16-35mm zoom
      • OK 60mm lens on P1 P45+
      • OK Sony 600mm f/4
      • Pretty good 16-35 FF zoom
      • Pretty good 90mm FF lens
      • Problematic 400 mm FF lens
      • Tilted 20 mm f/1.8 FF lens
      • Tilted 30 mm MF lens
      • Tilted 50 mm FF lens
      • Two 15mm FF lenses
    • Found a problem – now what?
    • Goals for this test
    • Minimum target distances
      • MFT
      • APS-C
      • Full frame
      • Small medium format
    • Printable Siemens Star targets
    • Target size on sensor
      • MFT
      • APS-C
      • Full frame
      • Small medium format
    • Test instructions — postproduction
    • Test instructions — reading the images
    • Test instructions – capture
    • Theory of the test
    • What’s wrong with conventional lens screening?
  • Previsualization heresy
  • Privacy Policy
  • Recommended photographic web sites
  • Using in-camera histograms for ETTR
    • Acknowledgments
    • Why ETTR?
    • Normal in-camera histograms
    • Image processing for in-camera histograms
    • Making the in-camera histogram closely represent the raw histogram
    • Shortcuts to UniWB
    • Preparing for monitor-based UniWB
    • A one-step UniWB procedure
    • The math behind the one-step method
    • Iteration using Newton’s Method

Category List

Recent Comments

  • JimK on Goldilocks and the three flashes
  • DC Wedding Photographer on Goldilocks and the three flashes
  • Wedding Photographer in DC on The 16-Bit Fallacy: Why More Isn’t Always Better in Medium Format Cameras
  • JimK on Fujifilm GFX 100S II precision
  • Renjie Zhu on Fujifilm GFX 100S II precision
  • JimK on Fuji 20-35/4 landscape field curvature at 23mm vs 23/4 GF
  • Ivo de Man on Fuji 20-35/4 landscape field curvature at 23mm vs 23/4 GF
  • JimK on Fuji 20-35/4 landscape field curvature at 23mm vs 23/4 GF
  • JimK on Fuji 20-35/4 landscape field curvature at 23mm vs 23/4 GF
  • Ivo de Man on Fuji 20-35/4 landscape field curvature at 23mm vs 23/4 GF

Archives

Copyright © 2025 · Daily Dish Pro On Genesis Framework · WordPress · Log in

Unless otherwise noted, all images copyright Jim Kasson.