3 ms·

But building optics that can properly illuminate those larger sensors within the confines of current smartphone designs already is impressive progress. As you

by usrusr 7y ago

But building optics that can properly illuminate those larger sensors within the confines of current smartphone designs already is impressive progress. As you said, those physics are hard. I'm no expert but I suspect that the physics would even be strictly impossible for a "proper" lens and they only get away with it because they can now apply corrections computationally which would require lot's of glass/space when done to light pre-sensor instead of to data post-sensor. This, I think, is the actual innovation: when you don't have to care for distortion or uneven brightness because you can fix that digitally, basic sharpness becomes much easier to achieve. A truly raw sensor readout from those modules would probably look like worse than a Holga picture. I fully agree with your assessment that this is a one-time free lunch. Looking forward to a computational compact that scales the same lens simplifications to a full frame sensor!

alkonaut 7y ago
Yes (more) computantional imaging is probably the next frontier. We already see the double and triple cameras (some times for depth info, some times for different FOV's) but I'm guessing N modules in some grid pattern + computation can emulate a much larger camera without making it much thicker like a single large sensor). I think I saw such a prototype if I remember correctly. If it works, it's also quite similar to that black hole imaging tech which is funny. I'd also be happy to get a phone with a detatchable camera module you could bring when you need it. So long as it's much smaller and cheaper than a camera, it would still be better than carrying a compact camera and a phone (which was the problem these good cameras solved in the first place).
- bitL 7y ago
  > N modules in some grid pattern Nokia 9 is like that.
- usrusr 7y ago
  Multiple sensors, stuff like Google "night sight" and all that fake bokeh are impressive. But compared to the benefits from liberating the optical path from the burden of minimizing distortion and vignetting they are little more than lipstick one a pig. Simplified optics improve the source signal, some hard trade-offs that defined camera design for more than a century have been blown wide open by a simple 2D convolution. Admittedly, multiple sensors also create more/better source signal to work with and unlike optical path simplification it's an improvement that can be repeated over and over again (just add more), but currently it's main importance is that it is something for buyers to see, without diving into the subjectivity-infested depths of image quality comparisons. I have no doubts that a bad multi-sensor would sell better than a good single-sensor.

by usrusr 7y ago

alkonaut 7y ago

Yes (more) computantional imaging is probably the next frontier. We already see the double and triple cameras (some times for depth info, some times for different FOV's) but I'm guessing N modules in some grid pattern + computation can emulate a much larger camera without making it much thicker like a single large sensor). I think I saw such a prototype if I remember correctly. If it works, it's also quite similar to that black hole imaging tech which is funny. I'd also be happy to get a phone with a detatchable camera module you could bring when you need it. So long as it's much smaller and cheaper than a camera, it would still be better than carrying a compact camera and a phone (which was the problem these good cameras solved in the first place).

bitL 7y ago
> N modules in some grid pattern Nokia 9 is like that.
usrusr 7y ago
Multiple sensors, stuff like Google "night sight" and all that fake bokeh are impressive. But compared to the benefits from liberating the optical path from the burden of minimizing distortion and vignetting they are little more than lipstick one a pig. Simplified optics improve the source signal, some hard trade-offs that defined camera design for more than a century have been blown wide open by a simple 2D convolution. Admittedly, multiple sensors also create more/better source signal to work with and unlike optical path simplification it's an improvement that can be repeated over and over again (just add more), but currently it's main importance is that it is something for buyers to see, without diving into the subjectivity-infested depths of image quality comparisons. I have no doubts that a bad multi-sensor would sell better than a good single-sensor.