The camera AI, that is. Last year I published an article reflecting on photography predictions I made in 2011 after I bought my Canon 7D. At the end of the article, I made new predictions about what would happen in the next decade. It’s been a little over a year and many of those things have already happened or will be happening soon. Let’s see how my predictions have progressed.
Global Shutter
Sony announced the A9 III in November and released it in March. It includes the world’s first full-frame global shutter image sensor that reads all of its data simultaneously. This means there’s no shutter warping of fast moving subjects and no limits on flash sync speed, both of which are a huge boon for action photography. The “compromises” of this technology are that the A9 III’s sensor is 24 MP (not 50 MP like the A1, but also no different than the A9 II) and that base ISO starts at 250 instead of 100. It has a tad bit more noise and a bit less dynamic range than its competitors due to its higher base ISO, but that generally doesn’t come into play at the high ISOs used for action shooting.
120 fps RAW
The A9 III can take 120 fps images RAW at full resolution with full autofocus and no image size restrictions until its buffer is full (1.6 seconds). My Z8 can also take 120 fps images, but they’re only JPEGs and they are scaled down to 11 MP. Capturing 120 RAW images every second creates a lot of data to push around, equivalent to a slow motion movie at between 6K and 8K resolution.
AI-based Noise Reduction
Canon announced the development of its flagship R1 camera in May. Details were vague, but Canon mentions deep learning and an image noise reduction function, noting specifically that it was “previously developed and improved as part of the software for PCs”. There are no other details, but this sounds like the same kind of AI-based noise reduction that DXO PureRAW and Lightroom can do, but in camera. Is it real time? I don’t know. It could be something that needs to be applied after the image is taken or might be limited to low frame rates or JPEG files, but it opens up possibilities. Canon plans to release the R1 this year and it will be very interesting to see how this works and whether it’s worth doing in camera or not.
Predictive Autofocus
This one isn’t quite realized just yet, but it’s on the way. The Sony A7RV and A9 III include an AI coprocessor that can estimate basic human skeletal structure and poses, even tracking their eyes when they turn their head away from the camera and back. The Canon R1 will be able to track a player even if someone passes in front of them and will also include an “Action Priority” mode that can detect a player in the scene shooting a ball and automatically focus on them. Again, the details are light, but it will be interesting to see how this works. I was thinking of something that would anticipate what direction the subject would move, but this is a start. Maybe it’s closer to what I imagined than I think.
The Speed of Progress
I definitely didn’t think this stuff would be coming so quickly. I knew Sony was working on global shutters but I expected them to start at the APS-C size since it’s smaller and thus cheaper and theoretically easier to produce. Sony really hit it out of the park with a full frame sensor whose only negative is that it only starts at ISO 250.
I really didn’t expect to see 120 fps RAW this soon. The Sony A1 was the king of high-resolution RAW (50 MP) at 30 fps and I assumed that increases to burst rate would happen incrementally from there. The Canon R3 already has, kind of, with the ability to take up to 50 RAW frames (24 MP) at 195 fps1. I assumed that the next leapfrog from any manufacturer would bump the rate to 60 fps and then maybe 120 fps a generation later. That’s two more generations of cameras and probably 4+ years away. I did not expect a quadrupling of RAW frame rate, without limitations, in a single generation.
Without knowing the details of Canon’s AI noise reduction and autofocus, I’m not sure if I’m surprised yet. If the noise reduction is similar quality to DXO PureRAW or Lightroom, it can be done automatically, doesn’t generate giant RAW files, and doesn’t eat the battery for lunch, it will be amazing. I think it will be really difficult for all of those things to be true. I definitely don’t think it will be able to do it at capture time or during bursts. Predictive autofocus is similar: it could be amazing or it could be limited to very specific scenarios. We’ll just have to see.
In any case, the future is coming faster than I thought it would, and that’s nothing but good for me. Think of all the features my next camera will have.
- It’s kind of a joke though as that’s only ¼ of a second of shooting ↩︎
Relatively Ambitious 