September 03, 2003

One More Expensive Camera

John Robb predicts that I will buy only one more expensive camera in my lifetime:

John Robb's Weblog: This last weekend my brother and I were discussing how long it would take for reusable digital cameras to become as inexpensive as calculators (from $1 to ~$125).  It's fairly clear that the feature set for digital cameras can't extend indefinitely, as has the personal computer (many people thought that PCs would go the way of calculators in the early 80s).  So, the question is: when does the market run out of gas (product differentiation) and move to pure price competition?  There clearly isn't much more headroom in megapixel growth (4 MP is fine for most purposes, particularly when the photos look fantastic on a $150 photo printer or on a 21" monitor).  I expect that differentiator to fade when we hit 7 MP in less than 2 years.

Next is storage.  Disk drives, ala personal storage devices (iPod and Archos), will clearly eliminate memory sticks and insertable memory soon.  10 gigabytes of storage will hold 10,000 photos.   The next phase will be digital video features and increased memory to hold those videos.  That race will end in the next five years with 1600x1200 XGA resolution and 500 gigabytes of memory (enough to hold ~5,000 3 minute clips).  Other items we will see on the way include solid state lens technology, larger viewing screens, and wireless connectivity (all of which have limited headroom for differentiation). 

Pure price competition is on the way and I suspect we will see the first reusable 7 MP digital camera available for under $20 in five years.

It's this way with every industrial revolution: some things that had been normal become really strange, some jobs that humans do become extremely valuable, more jobs that humans used to do vanish, and some things become really, really cheap.

It was not obvious to me a decade ago that cameras were one of those things that were going to become really, really cheap. It is now.

Posted by DeLong at September 3, 2003 03:25 PM | TrackBack

Comments

I have a 4MP camera at the moment, and I agree that I'll probably only need one more better one in the future.

But that doesn't account for a product's actual usuable lifespan (before it falls apart) and good old fashioned "I wants" (when something especially slick looking is released). I'm certainly looking forward to this theoretical ultra-camera in 2008, but I doubt that'll be the end of the story...

(ps when are disposable digital cameras due out?)

Posted by: stephen fulljames on September 3, 2003 03:47 PM

I have a 4MP camera at the moment, and I agree that I'll probably only need one more better one in the future.

But that doesn't account for a product's actual usuable lifespan (before it falls apart) and good old fashioned "I wants" (when something especially slick looking is released). I'm certainly looking forward to this theoretical ultra-camera in 2008, but I doubt that'll be the end of the story...

(ps when are disposable digital cameras due out?)

Posted by: stephen fulljames on September 3, 2003 03:48 PM

John Robb severely underestimates just how much more room there is for camera resolution and video memory use to grow.

For one thing, the human eye can resolve to better than 4,000 by 4,000 pixels, so 16 MP is probably a better upper limit; another thing to consider is that people will want to be able to zoom in on portions of pictures they've taken without losing resolving power (think Blade Runner!)

Between those two trends, and the utility of having better-than-32-bit color for video post-processing (DirectX 9 graphics cards already operate using 96-bit floating point internal accuracy), I doubt that Robb's predictions will pan out anytime soon. This sort of thinking reminds too much of the infamous "640K should be enough for everyone" comment.

Posted by: Abiola Lapite on September 3, 2003 06:46 PM

Why is it that when something is digital, people think that the only thing that matters for performance is how many megabytes (or megahertz, or mega-anything) it achieves? It's just a different way to store the same old information - nothing more and nothing less.

Newsflash: a high quality lens will still be expensive to manufacture, and contrary to what some people might thing, digital cameras still DO need lenses, just like film cameras. Maybe you will be able to buy a 7 MP camera for $20, but the picture quality will be no better than those disposable party cameras that currently sell for under 20 bucks. Having a billion megapixels still isn't going to give you stunning photos if you are using a crappy lens.

Posted by: huh? on September 3, 2003 07:31 PM

Huh? is correct. Cameras need lenses. Good quality lenses are expensive.

Film cameras are just boxes that hold film and a lens and have a few devices for focussing, setting exposure, etc. A digital camera incorporates the "film," but the lens remains an optical device. An N megapixel camera is a camera with better film than an an N-1 megapixel camera, but without good lenses the difference doesn't matter.

Posted by: Bernard Yomtov on September 3, 2003 07:59 PM

Right on those who realize that the lens is the missing piece of Robb's argument.

You can already buy an awesome 35 mm point and shoot camera that will meet all your needs for under $200 (I have the Yashica T4/T5 and it takes incredible photos). What Robb is describing is essentially the digital equvialent of that: a point and shoot snapshot camera with lots of storage space.

But for hobbists, that's not going to cut it. They're going to want an SLR digital camera that can mount their expensive Nikon or Canon lenses.

I would also like to point out that the current resolution of monitors will increase dramatically in the next few years. IBM has a monitor (currently used for medical imaging) which has the resolution of 4 1280x768 monitors. The resolution of paper (as Edward Tufte points out) is much, much higher than a 21 inch monitor. And I suspect that organic LCD displays are going to take us into the realm of 60 inch and larger displays on a regular basis. So there is plenty of room to move up on resolution.

Posted by: Luke Francl on September 3, 2003 08:12 PM

Ah. But what does nanotechnology mean for lenses?

:-)

Posted by: Brad DeLong on September 3, 2003 08:33 PM

"I doubt that Robb's predictions will pan out anytime soon. This sort of thinking reminds too much of the infamous "640K should be enough for everyone" comment."

Abiola, I agree with you that his timeline is too agressive, but I think the concept is correct--technology to be used directly by humans eventually bumps into the limits of humans.

Yeah there were silly comments like the 640K one, and that the world will need a few computers at most, but these comments made the mistake of not even considering future uses. When they proved to be ridiculous, many people instead adopted the opposite outlook where if you build the processing power the software developers and the gadget makers will come. In some sense this is already proving untrue, not as a hard limit, but as a gradual lowering of utility. For example, the popularity of $500 computers has already disproven Intel's claim a few years back that sub $1000 PCs would never catch on since there were going to be all these new applications that ate up cycles (Intel used to disparagingly call the sub $1000 machines 'sub zeros'.) Right now I only have 25 bucks of memory and the OS cannot seem to fully utilize it. In six years maybe the utility will be even lower and I will not know what to do with $10 of memory, even though that would give me 5x the amount I have today. In a dozen years perhaps the memory will be built onto the CPU adding a couple of bucks to the cost but sufficient for 99% of users.

Anyway, for $200 to have a nice tapeless DVD quality video camera, decent quality digitial still camera, and mp3 player will still be quite awesome. Right now to have all this costs easily over $1000.

I see there are many comments about the cost of the lens....perhaps a camera with a cheaper lens could be programmed to reduce distortions and give higher quality pictures? A film camera can't do that.


Posted by: snsterling on September 3, 2003 08:47 PM

Apart from anything else, the expensive part of any camera is the lens. The Nikon (chemical film) camera that I recently bought cost so much not because of the electronics or the shutter, but because of the big chunk of glass hanging off the front. The price of that big chunk of glass is not likely to change much in the immediate future.

Posted by: Sean Matthews on September 4, 2003 12:24 AM

"But what does nanotechnology mean for lenses?"

I suspect even with nanotechnology, getting optical precision is still going to be work. No biological lens is precise; eyes operate by integrating multiple samples, but producing a print such a process is likely to look more like experimental art than what we now call a photographic print.

4 MP is roughly the resolution of medium-resolution 35mm film; that isn't, in fact, enough for best-quality 8x10 prints; even wedding photographers use larger film, 60mm ("medium format"), which is roughly equivalent to 25 MP, for their best prints. 35mm has been way oversold to amatuer photographers; one of the best pieces of advice for most beginning serious photographers is still to use medium format.

Posted by: Randolph Fritz on September 4, 2003 01:03 AM

I think 4MP is a low estimate of the resolution of 35mm film, though as I understand it there is no clear agreement on the translation of film resolution to MP.

In any case I disagree that it is necessary to use medium format. Kertesz, Capa, Cartier-Bresson, Doisneau, and lots of others did just fine with 35mm.

Obviously larger film size is desirable, but it carries costs in terms of the weight and size of camera and lens, not to mention expense. The choice of film format is like the choice of any other tool. Different sizes are suitable for different jobs.

Digital enthusiasts should remember that film resolution is improving also.

Posted by: Bernard Yomtov on September 4, 2003 04:51 AM

First, when I read this post, I was a little suprised, and felt compelled to point out the obvious problem of lenses. That has been covered and the posters have done a good job, but upon reflection there is more...a lot more.
If PCs had stayed at a primitive level - imagine 1985 - then a PC would cost as much as a calculator. The reason it does not is because we have made improvements to PCs. Other people have mentioned resolution, so I will not delve into that except to reiterate that often times you will want to zoom in (Imagine taking a picture of your high school class of 500 people from the roof of your high school, and then 10 years later wanting to see what someone looked like- you would want the higher resolution).
But there are other things as well. Are your pictures as good as the best professionals? Probably not, and the reasons are many. Do they have a steadier hand, resulting in a crisper shot? Probably, but mostly it has to do with knowledge about the medium: Where is the light coming from, reflecting, what is the intensity, the range. With this type of action, what shutter speed do I need, what aperature gives me the proper depth of field? Already cameras help with many of these things, and with some things are better than the best professionals in terms of speed of decision making, but they have a long way to go on most of these factors.
And why does the digital photography medium have to try to mirror film photography? It actually has several areas where it is impossible to do completely different things. When shooting film, the film records the amount and quality of the light that accumulates in a certain area. Film is generally limited in range of the spectrum (either color OR infrared, etc.). Digital, on the otherhand could combine an entire range of spectrums, theoretically. It could do so while adding information that tracks the distance of a particular pixel( by sending out information that gets reflected back). Additionally, digital could keep track of the time that light strikes the sensor, making shutter speed calculations obsolete( The camera could remain open, later deciding the right light information to use by calculating the shift of the camera or the shift of the scene).
There are many other areas for improvements, and the short sightedness displayed by Robb and really does resemble the argument in the 80s regarding PCs.
As an aside, nanotechnology for lenses could be amazing, but when and if it comes in my lifetime it will have so many revolutionary impacts that to constrain it to talking about cameras makes no sense. It is like the one-panel-comic where the scientist's diagram on the blackboard includes the step "and then a miracle happpens".

Posted by: theCoach on September 4, 2003 06:30 AM

A cheap digital camera with perspective control, i.e., the image control you get with a large format camera by raising, swinging, and tilting the lens and/or film planes, would be something indeed. I'd be surprised if perspective control, like zooming, couldn't be implemented in the camera's software now. But somehow I doubt there will be much demand for this feature, and so I'll have to continue to smile through people's photos of tall buildings that look as if they're coming to a point, and nod politely when the pictures are favorably compared to those of Bernard and Hilla Becher's.

Posted by: Curtiss "You can have my Sinar F when you can pry it from my cold dead hands!" Leung on September 4, 2003 06:48 AM

Interesting responses on this thread.

I think the first point is that, yes, ultra-high professional quality stuff will still cost more, but consumer grade equipment really is going to get absurdly cheap and much better.

The second point, and one that I don't expect afficianados to accept very quickly, is that advances in sensor technologies (e.g, think 3 generations beyond Foveon) and computing power will soon start giving us cameras the go beyond the limitations of the relatively crappy and expensive lenses we're using today. Basically, they will have the same (or worse) lenses, but much, much better post-processing. We all know that the human eye is a pretty horrible optical device, and that our impressive visual abilities are more a function of some very fancy software on the back end. Well, I don't see much reason to doubt that this is what imaging manufacturers will really start going for in the near future. Perfect glass lenses will still be tough to grind, but computing your way around most of the known limitations of your current lens will become dramatically easier. I think this will take somewhat longer than the original proponent, but it will indeed happen.

I know that slide rules are extremely collectible these days because they are beautiful (albeit obsolete) instruments. I have no question that high-end non-digital camera gear will end up in the same situation, although I can't tell you when the cross-over point will happen.

Posted by: Jonathan King on September 4, 2003 07:56 AM

Brad DeLong's prediction: "Pure price competition is on the way and I suspect we will see the first reusable 7 MP digital camera available for under $20 in five years."

My predictions: In 5 years, one will be able to purchase a good 3 MP digital camera for $50-100. A 7 MP camera (which isn't even commercially available...at least in Best Buy or similar stores) will still be over $150.

Many people have commented on lenses being a problem to bring down in cost. (Brad DeLong mentioned nanotechnology. Hope he's not counting on nanotechnology for lenses in the next 5 years. :-))

I don't know much about lens problems (which has never stopped me from commenting on a subject before), but I'd be surprised if computer power couldn't eventually overcome problems with computing power.

Here's a website discussing various lens optical problems. I'd be surprised if we didn't eventually have the computing power to correct at least some of these problems. Chromatic dispersion seems particularly simple to correct with computing power...at least to someone who doesn't know anything about the subject:

http://www.math.ubc.ca/~cass/courses/m309-03a/m309-projects/le/intro.html

Posted by: Mark Bahner on September 4, 2003 08:07 AM

Oops, corrections:

1) Should have been, "...but I'd be surprised if computer power couldn't eventually overcome problems with *lenses*."

2) "Chromatic dispersion" = chromatic aberration, to people who know what they're talking about. Either way, the problem seems basically to be that the various colors of the spectrum (from red to blue/violet) aren't focused to the same point. Whatever one calls it, it seems to me that computer image analysis could fix it.

Posted by: Mark Bahner on September 4, 2003 08:16 AM

This article appears in BusinessWeek, Sept.8, 2003, p. 109:

Disposable Digital

Dispposable cameras go digital with Ritz Camera's Dakota Digital ($11). There's no LCD display and no ability to download your shots to your computer, but you can delete your last shot if you think you've blown it. You still have to return it to a Ritz store--it costs another $11 for prints and a photo CD--just like a single-use film camera.

Don Coffin

Posted by: Donald A. Coffin on September 4, 2003 09:12 AM

Jonathan King makes an interesting point with regards to post processing, but I think he underestimates the some of the difficulty. First, we have very little idea how good the single frame image that we get in our minds is. Our brain's processing of visual cues is very different than a single fram 2-d image. To start with, we can only focus on a very small area with good resolution. In addition, our visual system never dilineates a time period, or frozen image[ See my post above for this concept in a camera]. We are constantly, updating.
We generally do not have 'photographic' vision, we process objects. As an example watch an amateur try to draw real life pictures- they get the objects but the perspective and the shading has to be taught or studied. And object recognition is hard...
We have simple but interesting object recognition going on now, but the difference between our abilities and the current state of computers is at this point unknowable- we need to get into the range of true AI to understand this. It is possible that we will be able to make a pretty comprehensive catalog of objects and how they should be photographed, and that it would work fairly well, but it is not yet clear what exactly we will be able to do with it.
Let me give an example: Take a very good high resolution photo of a person's face, and a similar image taken from a lesser quality camera. Now, say we want to blow up the image as a close up where the pores on the face are recognizeable, or the scruff, etc. This look is used a lot in artistic portraits. In the high quality image we simply enlarge. But in the lower quality image when we enlagrge we get a blurred image and we cannot recognize individual pores or hairs on a face- so what do we do for post processing? This simply is not a problem for our visual system. We know those pores are there, but we cannot make them out, unless we move closer, something akin to enlarging. But with a frozen image we then need to either keep the object blurry, or interpolate what the pores of a person with such a look might have.
At this point it may no longer be photography( a post processor that does an admirable job of this should be close to being able to create photographic whole human beings from a description or a few visual cues, correcting this or correcting that.
As I ended my earlier post, yes, this stuff might be coming, but it is of significantly different character than qualifies someone to make assumptions that are being made. Clearly, the price for reasonably good consumer cameras is going to come down, but people seem willing to pay a certain amount to make their memories of a higher quality, and the goal of getting higher quality images might see diminishing returns, but that is by no means certain.

Posted by: theCoach on September 4, 2003 09:34 AM

Postprocessing shouldn't need any knowledge whatsoever of what the image is of. All it needs to know is how the lens in front of it distorts the image coming through, and the math to reverse those distortions (Rosalind Franklin in every snap).

I imagine a factory that makes lenses with only modest quality control, mounts them all into basic cameras, has each take a picture of a carefully designed test image, sorts them into blah-better-best, and then spends a night of computer power teaching each camera individually to correct for its particular lens.

Posted by: clew on September 4, 2003 10:00 AM

All of this assumes that Microsoft won't exercise its unchallenged, unpunished market power. I believe that we will get fair efficient markets only in our dreams.

Posted by: David on September 4, 2003 10:09 AM

"John Robb severely underestimates..." and I do agree with Abiola on this. Here they market these camera cell-phones, it is of course a huge advantage to be able to send the pictures directly via (3G-)cellphone over having to go to a computer and email the photos. I upgraded from my old 1M cam to a 3M one and got better optics and videoclip recording. Nice, but next I'll want a smaller cam, and then I'll want the same quality but inside a cell-phone, and then...

To me its a nice toy, but it doesn't make breakfast or change diapers or vaccuum-clean.

http://blogofpandora.blogspot.com/2003_09_01_blogofpandora_archive.html#106269544503035376

And what really should be discussed in this context is these surveillance cameras that soon will be *everywhere*. It's not only big brother either, think about the Riyhad bombings where the explosives where remote controlled by cell-phones. What would these guys be able to do with cell-phone cameras? I mean, with this rapid productivity growth, every small terrorist organisation will be able to build their own cruise missile.

Posted by: Mats on September 4, 2003 10:27 AM

Jonathan King makes an interesting point with regards to post processing, but I think he underestimates the some of the difficulty. First, we have very little idea how good the single frame image that we get in our minds is. Our brain's processing of visual cues is very different than a single fram 2-d image. To start with, we can only focus on a very small area with good resolution. In addition, our visual system never dilineates a time period, or frozen image[ See my post above for this concept in a camera]. We are constantly, updating.
We generally do not have 'photographic' vision, we process objects. As an example watch an amateur try to draw real life pictures- they get the objects but the perspective and the shading has to be taught or studied. And object recognition is hard...
We have simple but interesting object recognition going on now, but the difference between our abilities and the current state of computers is at this point unknowable- we need to get into the range of true AI to understand this. It is possible that we will be able to make a pretty comprehensive catalog of objects and how they should be photographed, and that it would work fairly well, but it is not yet clear what exactly we will be able to do with it.
Let me give an example: Take a very good high resolution photo of a person's face, and a similar image taken from a lesser quality camera. Now, say we want to blow up the image as a close up where the pores on the face are recognizeable, or the scruff, etc. This look is used a lot in artistic portraits. In the high quality image we simply enlarge. But in the lower quality image when we enlagrge we get a blurred image and we cannot recognize individual pores or hairs on a face- so what do we do for post processing? This simply is not a problem for our visual system. We know those pores are there, but we cannot make them out, unless we move closer, something akin to enlarging. But with a frozen image we then need to either keep the object blurry, or interpolate what the pores of a person with such a look might have.
At this point it may no longer be photography( a post processor that does an admirable job of this should be close to being able to create photographic whole human beings from a description or a few visual cues, correcting this or correcting that.
As I ended my earlier post, yes, this stuff might be coming, but it is of significantly different character than qualifies someone to make assumptions that are being made. Clearly, the price for reasonably good consumer cameras is going to come way down, but people seem willing to pay a certain amount to make their memories of a higher quality, and the goal of getting higher quality images might see diminishing returns, but that is by no means certain.

Posted by: theCoach on September 4, 2003 10:33 AM

I apologize for the double posts (although I really blame De Long's post mechanism).

Clew,
Not knowing much about lens theory, it would seem that some factors could be fixed without knowing what the pictures were of, but not most.
There are many different ways to measure the quality of lens. Sharpness, for example, cannot be fixed, or at least it is non-trivial to fully account for a difference in sharpness (just think of the extremes). I would imagine as well that these qualities are affected by aperature, fouc and distance information, and on zoom lenses, the focal length as well.


Posted by: theCoach on September 4, 2003 10:58 AM

I apologize for the double posts (although I really blame De Long's post mechanism).

Clew,
Not knowing much about lens theory, it would seem that some factors could be fixed without knowing what the pictures were of, but not most.
There are many different ways to measure the quality of lens. Sharpness, for example, cannot be fixed, or at least it is non-trivial to fully account for a difference in sharpness (just think of the extremes). I would imagine as well that these qualities are affected by aperature, fouc and distance information, and on zoom lenses, the focal length as well.


Posted by: theCoach on September 4, 2003 11:03 AM

Coach,

Interesting. I think that we will that some types of in-camera post-processing are easy, others impossible.

One clue might be what can be done in Photoshop and similar software already. For example, Photoshop permits perspective adjustment. This could be done in-camera I suppose, by swinging and shifting the sensors, though it's hard to see how it could be done automatically.

On the other hand, things like blurred images may simply be logically unfixable. This is really conjecture on my part, but it seems to me that to make a correction requires that somehow the "incorrect" image contain enough information to enable the "correct" image to be calculated. Is the information to do this there, and can it be extracted and processed reasonably by the camera-size computer that will be available in 2013?

Posted by: Bernard Yomtov on September 4, 2003 01:04 PM

theCoach: If the lens doesn't actually block rays of light coming through, and the camera has been shown a lot of test images and told what it 'should' have seen, then I think you're being too pessimistic. I may be too optimistic, but my optimism is based on some lab equipment that was really computationally stupid by current standards. Still effective: blurs in, useful pictures out.

...I've been looking for a well-illustrated site on X-ray diffraction, as an example of problems solved that way even when computation was expensive because we hadn't any other way to solve them. Haven't found any.

Posted by: clew on September 4, 2003 01:37 PM

"...I've been looking for a well-illustrated site on X-ray diffraction, as an example of problems solved that way even when computation was expensive..."

I don't think it takes much imagination to look at the picture on this site showing a lens with chromatic aberration...

http://www.math.ubc.ca/~cass/courses/m309-03a/m309-projects/le/intro.html

...and envision how a computer could be used to solve such a known lens defect. It's just a matter of pushing greens and reds up or down a certain amount.

In fact, if one didn't have the computing power to do the adjustments inside the camera, I could see how one could just have the adjustments done by software on a more powerful (e.g. desktop) computer.

As pointed out, one could just have the camera take a picture of a given target or set of targets, and then have the software adjust until the camera image matches the true target image.

Posted by: Mark Bahner on September 4, 2003 07:51 PM

For one, there are plenty of 7MP and above cameras available today. If you are looking for a non-pro one, Sony F828 is a very nice 8MP camera. If their pattern stays true in less than 6 months we will see that same sensor in their cybershot line with fewer features.
I believe; however, Roob and Brad are not talking about the ultimate in photographic equipment. For them, I have to say that a friend enrolled at MIT was testing, along with his whole Sloan first-year class, a prototype of a tiny (flash memory-card sized) point-and-shoot digital camera that they are hoping will retail for under $40.

Posted by: Con Tendem on September 5, 2003 03:43 PM

For one, there are plenty of 7MP and above cameras available today. If you are looking for a non-pro one, Sony F828 is a very nice 8MP camera. If their pattern stays true in less than 6 months we will see that same sensor in their cybershot line with fewer features.
I believe; however, Roob and Brad are not talking about the ultimate in photographic equipment. For them, I have to say that a friend enrolled at MIT was testing, along with his whole Sloan first-year class, a prototype of a tiny (flash memory-card sized) point-and-shoot digital camera that they are hoping will retail for under $40.

Posted by: Con Tendem on September 5, 2003 03:48 PM

Last year we gave $35 digital cameras to the entire entering class at SIMS. Here's the result:
http://www.sims.berkeley.edu/academics/courses/is202/f02/PhotoProject.html

Some visiting researchers from Nokia were so impressed they gave the entire entering class camera phones. Watch this space:

http://www.sims.berkeley.edu/academics/courses/is202/f03/phone_project/

Posted by: Hal Varian on September 21, 2003 09:13 AM
Post a comment