"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."
I agree, if I remember correctly Lytro had 1M pixles (??) during the 10Mpxl era. But as larger sensor appear (to my surprise - a sensor is a fixed area of Si!) to become cheaper to make I think this problem will go away.
Also, these are problems that companies like Apple excel at smashing. Apple disdains what the consumer thinks they want and they have a legion of fan boys to become early adopters and proselytize the masses. Incidentally, I think the founder of Lytro now works for Apple.
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.
The F3 is an amazing camera. The F5 great is you want to have to bother with electronics (I'm faster without the electronics, but to each his own). Digital SLRs are an example of entrenched technology that doesn't make sense.
DSLR don't make sense. The mirror in film SLRs was always a compromise [1] that doesn't exist once you have a digital sensor. It was a compromise that became popular, at least in 35mm. The DSLR was a legacy format that was "professional" only because camera manufacturers would put things like big focus motors on mirrorless cameras. Market forces slowed the adoption of mirrorless and gave the SLR a reprieve to their inescapable demise.
So in a sense, ya, the F5 is greater than any D-SLR. The F5 is the pinnacle of the SLR, an excellent solution in the film domain. Digital SLRs are compromised format for a solution no longer needed.
[1] The SLR lets you see what the film will actually see. +1. The SLR has mirror shake -1, and a film plane that's far from the last lens element -1. For small "sensors" like 35mm, it (usually) wasn't that big a problem (lightweight mirrors, geometrically smaller distances between lens and focal plane). For medium format, SLRs (like Hassleblad) were neck in neck with other solutions (Rollei, Fujifilm). For large format, I don't think there was any SLR made.
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.
You shouldn't get (be) downvoted for having an unpopular opinion. Unfortunately, I think this is a dangerous path we're on as a society.
But I do think you're wrong.
1. It's unlikely that FB doesn't sell your data if the buyer is willing to pay enough, have close enough connections. There's an asymmetry in your relationship with FB. You have no way of verifying that they behave in a way they say they do, while FB has huge incentives to behave in a way they say they don't.
2. Binning people is information about who they are. There are 2^33 people in the world. With 33 50-50 bins you can de-anonymize anybody. If you have an idea or a target in mind (say Italians, and then you bin by language), you can do much better than 23 bins.
For example, I could target my wife with personalized ads with, maybe, 5 bins - very unlikely nationality, unlikely job for a woman, unique set of languages. In fact, without knowing the nationality, job, or language, you can now probably target my wife fairly easily.
You example for Vancouver illustrated this. There's maybe a few thousand 18-25 year old security specialists that live in Vancouver. I'd guess less than 10k, certainly less than 100 000. A small group of people can manually comb through 100 000 profiles in a few days.
3. Given the spectacular security failures in this business, a mere claim of secure isn't good enough. Surely FB hires top talent, but they also have a history of epic failures or disregard for security to deliver early. If I can't trust Sony, a large tech (if not IT) company, to upgrade basic security after getting PWND the same way thrice, why should I trust FB? After all, Sony had great programmers before Zuckerberg was born.
4. Personalized ads necessitate a model of each individual. Most lay people never stop and think that that is the case, but most people are loath to have any profile of them being built and stored by anyone. Nevermind a company that has a (I believe a well earned) reputation for being sleazy.
5. You prefer personalized ads. Most people don't, especially when you point out that personalized ads necessarily means that a profile of them is being built by FB.
Most people prefer unobtrusive, lightweight irrelevant ads. Apple is not blocking personalized ads, they're changing their default settings to what it believes most of its consumers prefer.
I have never met an Android user say they prefer Android for the better ad or privacy experience. It's always price premium, feature premium, UI, and tightly locked down phone. In fact, I've converted many would be Android users by pointing out Google's more interesting practices (personally I'd get a librephone, but my original SE is still going strong).
I agree, if I remember correctly Lytro had 1M pixles (??) during the 10Mpxl era. But as larger sensor appear (to my surprise - a sensor is a fixed area of Si!) to become cheaper to make I think this problem will go away.
Also, these are problems that companies like Apple excel at smashing. Apple disdains what the consumer thinks they want and they have a legion of fan boys to become early adopters and proselytize the masses. Incidentally, I think the founder of Lytro now works for Apple.