5 ms·

DSLR manufacturers already have compact mirrorless to catch up with. Now only if the mirrorless moguls would recognize the potential of ehnancing their comparat

by ansgri 7y ago

DSLR manufacturers already have compact mirrorless to catch up with. Now only if the mirrorless moguls would recognize the potential of ehnancing their comparatively giant APS-C sensors with at least half the computational photography wits of modern camera phones, now that would be magnificent. 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.

dagw 7y ago
I wonder why traditional camera manufacturers have been so reluctant to leverage the advances in computational photography that the mobile phone world has embraced. I mean I get why they want to have neutral default settings for people who want to do their own editing, but why are their 'scene' settings so crap when compared to mobile phones. Is it simply because they're worried that a sufficiently powerful processor is going to draw too much power?
- willis936 7y ago
  Conservative management. What happens to those who don’t swim?
  - fock 7y ago
    I'm into taking photos to craft an image of reality. Myself. I couldn't care less for any AI crafting it's image of typical reality for me. I think that's quite set for anyone taking pictures for fun vs. admoney/instafame/...
    - ansgri 7y ago
      There's still such thing as physically correct image reconstruction, i.e. removing technical limitations via computational means. Super-resolution, focus stacking, HDR (proper one, without fake colors), highlight color preservation (try to photograph light source while preserving color while not underexposing everything around). 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.
      - fock 7y ago
        And you can do all of this with an ILC too and will profit from the fact that it has far less technical limitations. 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...
        ansgri 7y ago
        I was talking of compact ILCs specifically, that they would profit from this to become comparable with bulky 'pro' gear.
        fock 7y ago
        ah, but then if you look at µ4/3 it's already there (6k/30fps bursts with the electronic shutter for selection afterwards, pixelshift to improve resolution, ...)
- KineticLensman 7y ago
  I can't speak for the manufacturers but as a DLSR user I love the total predictability (exposure, DoF, focus, etc) when it is in manual mode. I really don't want the camera to try to anticipate the type of picture I am aiming for. This totally applies to scenes. Portrait, sports, landscape, etc are all just different permutations of shutter speed, aperture and ISO.
charrondev 7y ago
I’m not sure how it is at APS-C (I use a full frame DSLR) but I don’t think there’s much room in the workflow of a lot people using DSLRs for more computation I’m the camera itself. 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.
- ansgri 7y ago
  It's exactly about rising the performance of current inexpensive APS-C cameras to the level of professional full-frame models. 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.
- prolepunk 7y ago
  Panasonic has really good image stabilization for MFT systems where lens stabilization and sensor stabilization system work in tandem allowing to take shots with up to 1 second exposure handheld. Basically it's like having gimble inside the camera. 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.
jug 7y ago
> would recognize the potential 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.
- ansgri 7y ago
  > they capture much more information with their larger sensor sizes Most modern cameras are around 24 MPx irrespective of sensor size, this one is 3x12=36 MPx which is about the same total information content (or even less) given these 24 MPx have more bits per pixel. 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.

by ansgri 7y ago

dagw 7y ago

I wonder why traditional camera manufacturers have been so reluctant to leverage the advances in computational photography that the mobile phone world has embraced. I mean I get why they want to have neutral default settings for people who want to do their own editing, but why are their 'scene' settings so crap when compared to mobile phones. Is it simply because they're worried that a sufficiently powerful processor is going to draw too much power?

willis936 7y ago
Conservative management. What happens to those who don’t swim?
- fock 7y ago
  I'm into taking photos to craft an image of reality. Myself. I couldn't care less for any AI crafting it's image of typical reality for me. I think that's quite set for anyone taking pictures for fun vs. admoney/instafame/...
  - ansgri 7y ago
    There's still such thing as physically correct image reconstruction, i.e. removing technical limitations via computational means. Super-resolution, focus stacking, HDR (proper one, without fake colors), highlight color preservation (try to photograph light source while preserving color while not underexposing everything around). 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.
    - fock 7y ago
      And you can do all of this with an ILC too and will profit from the fact that it has far less technical limitations. 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...
      - ansgri 7y ago
        I was talking of compact ILCs specifically, that they would profit from this to become comparable with bulky 'pro' gear.
        fock 7y ago
        ah, but then if you look at µ4/3 it's already there (6k/30fps bursts with the electronic shutter for selection afterwards, pixelshift to improve resolution, ...)
KineticLensman 7y ago
I can't speak for the manufacturers but as a DLSR user I love the total predictability (exposure, DoF, focus, etc) when it is in manual mode. I really don't want the camera to try to anticipate the type of picture I am aiming for. This totally applies to scenes. Portrait, sports, landscape, etc are all just different permutations of shutter speed, aperture and ISO.

charrondev 7y ago

I’m not sure how it is at APS-C (I use a full frame DSLR) but I don’t think there’s much room in the workflow of a lot people using DSLRs for more computation I’m the camera itself. 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.

ansgri 7y ago
It's exactly about rising the performance of current inexpensive APS-C cameras to the level of professional full-frame models. 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.
prolepunk 7y ago
Panasonic has really good image stabilization for MFT systems where lens stabilization and sensor stabilization system work in tandem allowing to take shots with up to 1 second exposure handheld. Basically it's like having gimble inside the camera. 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.

jug 7y ago

> would recognize the potential 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.

ansgri 7y ago
> they capture much more information with their larger sensor sizes Most modern cameras are around 24 MPx irrespective of sensor size, this one is 3x12=36 MPx which is about the same total information content (or even less) given these 24 MPx have more bits per pixel. 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.