3 ms·

I think cameras are done. And not because of cellphones. I think Lytro did them in. First a cop out to what I'm about to say. My arguments are specifically for

I think cameras are done. And not because of cellphones. I think Lytro did them in. First a cop out to what I'm about to say. My arguments are specifically for geometric optics as used in photography. Also, my argument is in the limit that (non-optical) technology gets better. Also, while I took (and loved) many optics classes at university a lifetime ago, I ultimately did not become an optical engineer. I'm sure there are many nuances in my arguments bellow. But largely they should be correct. The Lytro camera is a camera that records the angle of incidence of light rays. If you have the angle of incidence of rays then the entire problem of (geometric) lens design is finished. Everything a lens can do can be done in software. Think of Lytro as software defined radio. With a light field camera you can design the physical lens with a single objective: f-stop. Gather as much light as possible. You can adjust everything else in software. This is like SDR where (I guess) you optimize the antenna for bandwidth and noise. By adjust I don't mean "adjust" as in apply AI or heuristic to make it appear as if you've achieved a goal. This is what your cellphone does. I actually change the lens. Since you have the ray's angle of incidence you can roll the rays back to before they reached the camera lens, and then put your own, software defined, lenses. This is probably not how you would implement the algorithm, but it illustrates the point. Therefore you can change focal length, aperture, break physical limits (ridiculous index of refractions with no dispersion. infinitely thin lens). Back to reality, what does it take to achieve all of this? Optically the Lytro all ready did all of this. It failed, but for the reasons that an Apple product would hit it out of the ballpark - implementation. With better software/computer power the product would be easier more convenient to use. With bigger, better sensors (to better capture the ray's angle) the camera would be able to emulate more lenses. Therefore, 20th century type photography is dead. A modern photographer could just take a camera out, really take a shot without minding much about aiming it [1], completely ignore light and focus, and everything else could be done in software. Frankly I think this will kill photography - but honestly, I think cell phones have killed most non-art, non-professional photography already. Amateurs are far worse (i.e. boring) photographers today then 30 years ago. That's a judgement call, sure, but that's what I enjoy. [1] As long as the subject is captured, it won't really matter. You can pan in post processing. You can even adjust for perspective [2]! Just roll the rays outside the physical lens, and move your software lens to where you want it. [2] Perspective - the size of elements of the scene wrt to each other - is strictly about where the camera is wrt to the subject. It has nothing to do with the lens. Lazy photographers make the mistake of "zooming" into the subject to get close - but the subject is as flat. In software, you can move the lens, and therefore even change perspective.

wenc 6y ago
I remember Lytro. Being able to adjust focus in post was an exciting feature, but the cameras themselves were expensive and had lower quality sensors than the cameras of that time (ca 2012). As you say, the implementation failed but it demonstrated that the technology was feasible. Now the IP is with Google. I'm just waiting for the day when someone perfects light-field technology for cell phone cameras.
- bgrsgeg 6y ago
  Lytro failed, I assume, for the same reasons that companies like Apple makes winners out of failed ideas. When it came out, I nerded out and wanted one so badly. Luckily I couldn't afford one. From what I read was that the software was slow, and it was a pain in the butt to go back and re-frame or refocus a shot. People don't do post-processing for mere white balance [1], why would they bother for focus, focal length, depth of field, pan, and perspective? On the other hand, it means that high end gear is done. Once a well executed light field camera comes to market professionals will flock to it. They have to do post-processing anyway so they'll adopt the camera that is smaller, lighter, single insanely fast lens. Most importantly perhaps, a camera where they don't have to wrestle the bloody auto-focus (the kid! focus of the running kid on the edge of the shot!). Better sensors and faster CPUs and digital cameras as we know them are dead. [1] That's why I like film, preferably B&W. Take the time to do the right shot. Accept the results.
analog31 6y ago
I do some optics design. The Lytro was exciting. Reinforcing what you say, when a light ray hits a pixel on a conventional sensor, the sensor only knows where the pixel is located on the sensor. Ensuring that each pixel corresponds to a specific location in the scene being photographed is the lens design problem in a nutshell. But a ray has both a position and a direction vector. If you can measure the complete ray vector -- its position and direction at the same time, then you can compute where it came from, even if the lens is not perfectly designed. The light field sensor relaxes the lens design problem considerably. An earlier technology called a Shack-Hartmann sensor, similarly measures the full light field at once. It's used for things like measuring the wavefront quality of lasers. There is still a problem, which is that you have to assign a physical sensor pixel to each point in a 4-d space (2 position and 2 direction coordinates), which means that you have to give up some of the spatial resolution of the sensor. This makes it hard to compete in a market that is clamoring for more and more megapixels. Instead, the sales pitch has to be more nuanced, as in comparing the actual image quality of the pictures, which is a harder sell when the quality of pictures from cell phone cameras is pretty damn good.

wenc 6y ago

I remember Lytro. Being able to adjust focus in post was an exciting feature, but the cameras themselves were expensive and had lower quality sensors than the cameras of that time (ca 2012). As you say, the implementation failed but it demonstrated that the technology was feasible. Now the IP is with Google. I'm just waiting for the day when someone perfects light-field technology for cell phone cameras.

bgrsgeg 6y ago
Lytro failed, I assume, for the same reasons that companies like Apple makes winners out of failed ideas. When it came out, I nerded out and wanted one so badly. Luckily I couldn't afford one. From what I read was that the software was slow, and it was a pain in the butt to go back and re-frame or refocus a shot. People don't do post-processing for mere white balance [1], why would they bother for focus, focal length, depth of field, pan, and perspective? On the other hand, it means that high end gear is done. Once a well executed light field camera comes to market professionals will flock to it. They have to do post-processing anyway so they'll adopt the camera that is smaller, lighter, single insanely fast lens. Most importantly perhaps, a camera where they don't have to wrestle the bloody auto-focus (the kid! focus of the running kid on the edge of the shot!). Better sensors and faster CPUs and digital cameras as we know them are dead. [1] That's why I like film, preferably B&W. Take the time to do the right shot. Accept the results.

analog31 6y ago

I do some optics design. The Lytro was exciting. Reinforcing what you say, when a light ray hits a pixel on a conventional sensor, the sensor only knows where the pixel is located on the sensor. Ensuring that each pixel corresponds to a specific location in the scene being photographed is the lens design problem in a nutshell. But a ray has both a position and a direction vector. If you can measure the complete ray vector -- its position and direction at the same time, then you can compute where it came from, even if the lens is not perfectly designed. The light field sensor relaxes the lens design problem considerably. An earlier technology called a Shack-Hartmann sensor, similarly measures the full light field at once. It's used for things like measuring the wavefront quality of lasers. There is still a problem, which is that you have to assign a physical sensor pixel to each point in a 4-d space (2 position and 2 direction coordinates), which means that you have to give up some of the spatial resolution of the sensor. This makes it hard to compete in a market that is clamoring for more and more megapixels. Instead, the sales pitch has to be more nuanced, as in comparing the actual image quality of the pictures, which is a harder sell when the quality of pictures from cell phone cameras is pretty damn good.