Also, for me personally, the next waypoint for the phones should be the development of compact telephoto lenses -- it seems only Huawei got that right with their P30Pro.
I know if I want a _great_ picture I
- shoot raw. - shoot on manual mode (often still with auto exposure). - put the photo into Lightroom.
With a huge sensor and a fast lens there’s no need for a lot of the computational photography phones are doing:
- “portrait mode” - unneeded when you have a fast lens and can get more real bokeh than you could ever need. - “night mode” - with modern sensors you can get usable images even at 6400/12800 ISO. This lets you get a solid image even at a reasonable, handheld shutter speed.
The real improvement other than sensor dynamic range and pixel count has been in autofocus. I still miss some some shots with my Canon because I didn’t react fast enough to nail my focus.
As an amateur photographer I'd never take with me a mammoth gear that is some Canon 5D mk3 w/ 70-200 F/4, yet I'm somehow pretty sure that some good processing would allow a compact like Sony a6000 w/ 18-135 F/5.6 get close to its performance in 90% of cases.
The thing is, I sometimes can do that in Lr/Ps, but it's just much more work than I'd like to do for most of the photos.
Panasonic 25mm f/1.7 (50mm f/3.5 equivalent full frame) is under $200 and is an excellent portrait lens.
With initial release of GH5 autofocus was pretty bad, but with the last firmware upgrade apparently it's quite good now.
I however cant yet justify upgrade from GH4, which doesn't have sensor stabilization.
Oh I'm sure they've recognized the potential all right. But these smartphone manufacturers have massive software development departments with heaps of funds poured into AI research for multiple reasons other than this specific field, which is now really starting to benefit them. It's a different culture.
And even if we assume a camera manufacturer would have the finances to build this culture on their home turf (a huge assumption to make), they then run into the next problem that their sensors are too big! They can't compete as strongly here because they capture much more information with their larger sensor sizes and even with tiny smartphone sizes, you need laptop level performance on a ~7-10 nm transistor scale and dedicated AI hardware to pull it off without destroying battery life. They're so far behind both in know how and difficulty of the problem that I'm not sure how they'll ever be able to do it in the foreseeable future at least with full frame cameras.
Another point in case is that one could implement more in-camera techniques for collecting raw data for further offline processing, like focus stacking or aperture simulation which today require external hardware and a decent amount of skill.
Basically, make images that are currently possible on $3000+ hardware, on $600 hardware.
There's also such thing as perceptually correct image rendering, but that's a whole another can of worms (image appearance models and the like), it isn't scientifically ready just yet.
Basically most of the really "cool" applications center around using the electronic shutter which up to 2-3years ago none of the ILC-companies dared to offer for the masses because it has some shortcomings which the "hey I can finally shoot in the dark"-crowd might not notice/care about, but people buying an ILC will, even if they're not technical enough to understand that it's their own problem...
This totally applies to scenes. Portrait, sports, landscape, etc are all just different permutations of shutter speed, aperture and ISO.