4 ms·

I don't want that versatility if it's going to compromise the integrity of the image just because some AI was trained on what the average yokel considers a "goo

I don't want that versatility if it's going to compromise the integrity of the image just because some AI was trained on what the average yokel considers a "good picture" I want the optics and processing to produce the most detailed, widest-gamut, highest-dynamic-range image that is possible, give it to me in a raw form, and then allow me to shape the image to whatever I need it to be AI R&D takes away from optics/processing R&D, and those are what should be getting 100% of the R&D effort

TaylorAlexander 3y ago
Right so maybe you aren't the target customer. > AI R&D takes away from optics/processing R&D Maybe it does, maybe it doesn't. I don't know what the engineering structure of the company is like. I just disagree that the goal or purpose of computational photography is to skimp in other areas. The G9ii has a 25 megapixel micro four thirds image sensor but it can take 100 megapixel photos using an automated process where the image sensor is physically shifted around (thanks to in-body image stabilization capability) while multiple exposures are taken and then combined using computational techniques. But that is certainly not as an alternative to producing a 100 megapixel micro four thirds sensor, which would have very small pixels and thus poor noise performance among other things. On the G9ii, the computational features purely extend the function of the system. You could argue maybe they would have gone to 30 megapixels or 35 if they did not have this alternative, I don't know. If you want the absolute best possibly image quality, you simply will not want a pocket action camera. But all cameras will adopt more computational techniques and I don't think it's correct to say they are doing so because they wanted to skimp in other areas. Action cameras already skimp on lens quality to get a pocket form factor, not because they are adding AI. Additionally on cell phones, the form factor has always been the reason for low quality lenses and small sensor size. AI clearly has a place in extending the capability of a camera system with other physical constraints. Perhaps some companies will use that as an excuse to use cheaper optics than what is possible, but I would argue we cannot infer they have done that simply because computational photography is used. Computers are simply cheap and these techniques will get added because they have zero or very low BOM cost impact.
- pdntspa 3y ago
  And that just begs the question of why... if we are talking about the incredibly-contemptible cop-out of "the average consumer" than this 100 megapixels is literally for the marketing bulletpoint. 25 megapixels is more than enough for those clowns. Again, we can figure out better optics and processing without having to computationally invent details that didn't exist. We are seeing the same thing happen with DLSS, new games are coming out with frankly unacceptable performance and the cop-out is "just use DLSS"
  - TaylorAlexander 3y ago
    > 25 megapixels is more than enough for those clowns. I find it irritating to talk to people who speak this way, just FYI. > Again, we can figure out better optics and processing without having to computationally invent details that didn't exist. Computational photography is not just inventing details that didn't exist. The 100 megapixel mode on the G9ii physically shifts the sensor a minute amount to collect details which are there but fell in between the pixels on other exposures, and then it computationally combines multiple frames to reconstruct the details. I want this functionality because I print my photos in large format and the printer is capable of reproducing MUCH more detail than my current 21Mp camera can. > we can figure out better optics and processing There will always be a limit to optics, and "processing" is always going to be used. Whether a marketing department calls that processing AI or not doesn't matter. Every camera has a manufacturing budget to fit a target price, and that will always mean trade offs and choices made in lens quality. Spending on "processing" will always be part of the calculation. Even on a high end micro four thirds camera, there is just a limit to what you can do even with expensive lenses, and computational techniques are going to be used to optionally extend the capability.
    - pdntspa 3y ago
      Are you trying to print billboards? Absolutely nothing about large-format printing is 'mainstream'. Sensor fuckery is still a hack that can't do many types of shots (any kind of quick movement for example) while leading people to believe you can. That's the very definition of stuff you should be using specialized hardware for. Are they going to incorporate a 10+ bit sensor and at least an sRGB color space too? Otherwise it's like putting 16gb of VRAM on a geforce 1060 I've done large format printing, you need at least those things in place for the print to come out accurately. > I find it irritating to talk to people who speak this way, just FYI. I find the gentrified, milquetoast corpo-speak that seems to be the norm here to be just as annoying

TaylorAlexander 3y ago

Right so maybe you aren't the target customer. > AI R&D takes away from optics/processing R&D Maybe it does, maybe it doesn't. I don't know what the engineering structure of the company is like. I just disagree that the goal or purpose of computational photography is to skimp in other areas. The G9ii has a 25 megapixel micro four thirds image sensor but it can take 100 megapixel photos using an automated process where the image sensor is physically shifted around (thanks to in-body image stabilization capability) while multiple exposures are taken and then combined using computational techniques. But that is certainly not as an alternative to producing a 100 megapixel micro four thirds sensor, which would have very small pixels and thus poor noise performance among other things. On the G9ii, the computational features purely extend the function of the system. You could argue maybe they would have gone to 30 megapixels or 35 if they did not have this alternative, I don't know. If you want the absolute best possibly image quality, you simply will not want a pocket action camera. But all cameras will adopt more computational techniques and I don't think it's correct to say they are doing so because they wanted to skimp in other areas. Action cameras already skimp on lens quality to get a pocket form factor, not because they are adding AI. Additionally on cell phones, the form factor has always been the reason for low quality lenses and small sensor size. AI clearly has a place in extending the capability of a camera system with other physical constraints. Perhaps some companies will use that as an excuse to use cheaper optics than what is possible, but I would argue we cannot infer they have done that simply because computational photography is used. Computers are simply cheap and these techniques will get added because they have zero or very low BOM cost impact.

pdntspa 3y ago
And that just begs the question of why... if we are talking about the incredibly-contemptible cop-out of "the average consumer" than this 100 megapixels is literally for the marketing bulletpoint. 25 megapixels is more than enough for those clowns. Again, we can figure out better optics and processing without having to computationally invent details that didn't exist. We are seeing the same thing happen with DLSS, new games are coming out with frankly unacceptable performance and the cop-out is "just use DLSS"
- TaylorAlexander 3y ago
  > 25 megapixels is more than enough for those clowns. I find it irritating to talk to people who speak this way, just FYI. > Again, we can figure out better optics and processing without having to computationally invent details that didn't exist. Computational photography is not just inventing details that didn't exist. The 100 megapixel mode on the G9ii physically shifts the sensor a minute amount to collect details which are there but fell in between the pixels on other exposures, and then it computationally combines multiple frames to reconstruct the details. I want this functionality because I print my photos in large format and the printer is capable of reproducing MUCH more detail than my current 21Mp camera can. > we can figure out better optics and processing There will always be a limit to optics, and "processing" is always going to be used. Whether a marketing department calls that processing AI or not doesn't matter. Every camera has a manufacturing budget to fit a target price, and that will always mean trade offs and choices made in lens quality. Spending on "processing" will always be part of the calculation. Even on a high end micro four thirds camera, there is just a limit to what you can do even with expensive lenses, and computational techniques are going to be used to optionally extend the capability.
  - pdntspa 3y ago
    Are you trying to print billboards? Absolutely nothing about large-format printing is 'mainstream'. Sensor fuckery is still a hack that can't do many types of shots (any kind of quick movement for example) while leading people to believe you can. That's the very definition of stuff you should be using specialized hardware for. Are they going to incorporate a 10+ bit sensor and at least an sRGB color space too? Otherwise it's like putting 16gb of VRAM on a geforce 1060 I've done large format printing, you need at least those things in place for the print to come out accurately. > I find it irritating to talk to people who speak this way, just FYI. I find the gentrified, milquetoast corpo-speak that seems to be the norm here to be just as annoying