It being Nature group is no guarantee of success or high impact. And we have no idea if they are in it for the long haul or if they'll bail in a few years if the uptake is low. I'd just wait a couple of years and see what happens to it before submitting a paper there. Meanwhile, PLoS has a good impact factor, large readership and doesn't have a limit on the number of accepted papers so that's a better option for now.
External harddrive. Second harddrive in different location and/or a flickr account so you don't have it only at home.
Needless to say we took thousands of pictures with two cameras. [...] These are obviously images that I want to keep for my life.
Now, seriously? I have photography as a hobby too, and in the very best case about one in ten shots are actually worth saving, and that proportion drops the more shots you take. Most pictures end up being crap - that's not an indictment, just a fact. Most shots Adams or Bresson took ended up as crap too. You have perhaps two hundred pictures, tops, that you or anybody else would care about.
Chances are you're never ever going to look at most of those thousands of shots ever again, and your kids will simply throw it all away unseen in the far future when they're cleaning out your belongings. Problem is, if you haven't edited the collection and thrown away all the thousands of duds, they will end up throwing away those good, important images along with the rest.
So, best approach: go through and delete all the crap, all the duplicates, all the technical duds. Then delete all the ho-hum images. Aim for, say, two hundred images to save, or better, fifty. No matter how eventful a vacation, you don't have more than fifty great places or events to record. And the more pictures you keep, the more you dilute the impact of the good images.
Anything that needs me to manually adjust the glasses is a no-go for me. Normal progressive lenses work just fine already, so I don't really see the point of making it complicated and intrusive. If I want to improve my eyesight I'd much rather get surgery for my shortsightedness and astigmatism, then use progressives with uncorrected upper area. Either way I don't have to mess with my glasses just because I want to change my viewing distance.
Meaning that if you're not going downtown you're almost certainly going to need to make a transfer. Good luck going east or west or around downtown.
Hub-and-spoke designs can be really frustrating. The problem is utilization, though: the town center is by far the most common destination so it's the one that can support most traffic. Even if most people want to go from one peripheral point to another, they all want to go between different points. A normal bus line would be too little used to be realistic.
You can't really adjust capacity all that much by spacing out the vehicles; people will only want to wait for so long to take a bus, and don't like having to plan their day around a bus time table. And you can't adjust the vehicle capacity too much either. A ten-passenger bus costs almost as much to run as a 50 passenger one. Driverless vehicles would be able to change that situation, but we're not there yet.
In a way it's a shame that it happened, because it only enabled the Japanese to continue limping along with their teeth-gnashing archaic writing system rather than simply adopting one of the very efficient, superior, and easily computable 38-character phonemic syllabary scrips that EVERYONE JAPANESE PERSON ALREADY KNOWS ANYWAY.
Armchair discussions on language efficiency, convenience and so on are really fun. Plenty of people argue that Japanese would be better without kanji. But the actual results is what matters.
We're collectively lazy, and our languages tend to become reasonably efficient over time. You can see how word length roughly corresponds to usage frequencies, how phoneme distribution and spelling tends to strike an efficient balance between brevity and redundancy so errors are minimized. None of this is consciously decided by anyone; instead people tend to gravitate over time toward usage patterns that minimize errors, redundancy and effort.
Japanese has, as you say, not just one but two syllabaries in addition to kanji, and has had them readily available for more than a millennium. A century or so ago there was an official drive to reduce the usage of kanji, with an eye towards abolishing them altogether. There has been lots and lots of time, and lots and lots of opportunity to move away from kanji in other words. So what happens?
People are using more kanji today, not less. Not from a sense of tradition, or from some official decree, but because they like to do so. They add expressiveness, and are another resource for punsters and jokers. The culture and education department recently acknowledged this trend and added another two hundred characters to the list that people are required to study at school.
Part of the reason use is increasing is because of electronic communication. Many of the drawbacks of kanji - it's difficult to remember how to write a rarely used character, for instance, or remember a rare pronunciation - are mitigated with cellphones and computers. It's much, much easier to recognize a character than two write it by hand, and with modern input methods that's all you need. And an unfamiliar word is easy to look up directly on the same device you're reading it on.
People are voting with their feet, as it were, and moving towards more kanji, not less.
[...]any time a copy needs to be made to backup or test, it takes an eternity because there are literally over 10,000 files on the disk.
How about using rsync? That should cut down a lot on the copy time. I haven't tried it myself, mind you; my world is still small enough that a straight copy takes very little time.
Well, the passage is an aside set in the middle of a sentence about farmers. Looks pretty clear to me, but since it's not clear to others I will try to be a bit more mindful about such things in the future. It kind of sucks when your argument misses the point because your text ends up hard to understand.
Yomiuri Shinbun just ran a series of articles looking at the current state of agriculture, using official figures as a base. Here is one relevant piece in English (warning: articles disappear offline after a few days): http://www.yomiuri.co.jp/dy/business/T101130005037.htm
Here's the relevant part:
The population depending on agriculture as their main revenue source has declined to about 2.6 million in 2010, one-third of the figure recorded 30 years ago, according to a census of the farming and forestry industries.
The statistics also show the farming population's average age rose to nearly 66 this year, with 1.6 million--more than 60 percent of the total agricultural population--aged 65 or older. A government official anticipates 100,000 people will quit farming annually in the years to come.
You may get different statistics if you count farmland owners rather than farmers. A lot of the uncultivated land is plots that have been inherited by people who live and work in the city. They live too far away to farm the land on one hand, and they don't want to sell the family land on the other, so it ends up sitting unused.
I think, perhaps, we might do well to understand what is driving them to that rather expensive farming method: lack of space.
Nope. There's plenty of farmland around in Japan; a fair amount of high-quality farmland lies fallow, in fact, from a lack of interest in using it. The reasons for the interest in factory farms (the indoor farming you've seen) and robotics are somewhat different. It's potentially much more effective and with higher-quality yield than open-field farming, where each plant gets the optimal amount of light and nutrients, no pests and no pesticides, and the same results no matter what the season or weather. And you can have small-scale farms that are still efficient, so a large restaurant could actually grow some of their own produce to the exact standards they want, all year round.
It would take fewer people to work such a farm too, and even more so with robots. The major reason for fallow land in Japan is the lack of people willing to be farmers in a distant rural area - the average age is already above retirement age - so this would be a way both to reduce the number of people needed, and move the industry to where people want to live and work.
This is what I meant about overselling this idea. A GPU is not an alternative to a cluster. The practical speedup you get is 2-3 times a normal dual-core cpu (a bit more for specific problems, less for others). A GPU is in other words very roughly equal to adding another 2-core cpu to your system, which makes the cost-benefit tradeoff clearer: cheaper node, but longer development time. If you need a cluster for your problem you're going to be using MPI no matter what*; one or two GPUs will make no material difference.
What I am talking about is clusters equipped with GPUs at each computing node. At which point you need to choose between spending the extra time developing for those GPUs; spending extra time on longer simulations without them; or spending extra money to get access to more computing nodes. And what I see (personal experience only, so not very reliable) from the clusters I have access to is that most researchers skip the GPUs in favour of the other two solutions today.
* Actually, some clusters can handle threaded applications across nodes, and that can be an alternative. But many massively parallel problems are really better expressed with something like MPI anyhow, and MPI is close to a true distributed processing standard. Use MPI on linux and your code is fairly easy to port between machines.
...even if you spent two months programming it, to gain two weeks of simulation time, those two weeks gained, are going to be used over and over and over in the future by countless users for years to come.
We're talking supercomputers, not pc:s. A lot of the software used there really is written for one single task or one single project. Once the project is over and the original user is done with it, it's never used again. While you may want to do something similar at a later date your new task is different enough - and the new machine you use is different enough - that you'll need to redesign the application. And if you were not part of the previous team, understanding and adapting their code (never mind finding the code in the first place) to your machine may well be more work than simply developing your own software from the start.
So the relevant measure is the total time spent during the lifetime of that project, and that means development time really does become a major factor to take into consideration.
GPUs are indeed an inexpensive way to boost speed in some cases. But they have been rather oversold; while some specific types of problems benefit a lot from them, many problems do not. If you need to frequently share data with other computing nodes (neural network simulations come to mind), then the communications latency between card and main node eats up much of the speed increase. And as much of the software you run on this kind of system is customized or one-off stuff, the added development time in using GPUs is a real factor in determining the relative value. If you gain two weeks of simulation time but spend an extra month on the programming, you're losing time, not gaining it.
Think about it this way: GPU's are really the same thing as specialized vector processors, long used in supercomputing. And they have fallen in and out of favour over the years depending on the kind of problem you try to solve, the relative speed boost, cost and difficulty in using them. The GPU resource at the computing center is used much less than the general clusters themselves, indicating most users do not find it worth the extra time and trouble to use.
It is a good idea, but it's not some magic bullet.
And yet we are able to navigate and interact with our environment with a high degree of precision.
Yes, we are. Our vision system is pretty successful when you look at how we actually use it in the real world. We don't actually need to know the precise distance to things; what we want to know is rather direction and time to impact and similar and we're really, really good at that (look up tau-margin estimation for instance). Though note that with a human-level vision system you would still need a lot of those sensors you talk about. Our vision system absolutely depends on proprioception to figure out where we are in the world and compensate for our own movements; we need separate dead-reckoning systems and (again) a lot of experience to be even somewhat correct about our movements over large distances and so on.
But I wrote this in reply to a poster that seemed to believe we humans are actually better than Kinect at the specific vision tasks it's built to do. Too many people seem to believe that the mammalian vision system is inherently great, at whatever tasks we imagine, and that if we could only make something like it our machine vision problems would be solved. That is simply not the case.
The "good ol' brain" does a fairly crappy job, actually. 3D vision systems like these tend to perform quite a bit better than we do. And we only do as well as we do because we can use a lot of indirect clues based on our long experience with a 3D-world - we know how big stuff normally is, for instance, so we can judge distance from size. Mess up those clues and we completely lose it.
And even with good clues we don't actually measure distance well. Have somebody place items on a parking lot or some place like that, then try to guess the distances. Not going to be very accurate. Try to estimate distance vertically rather than horizontally and you'll do even worse; you have fewer clues and less experience to fall back on.
Because taking apart a screwdriver is such an enriching experience.
Well, it is a completely different drink when you down the vodka neat then drink the orange juice as a chaser. The difference may be lost on many four-year olds I suspect, though.
Wbhy would he be punished? Memorizing the answer order isn't unethical or cheating or anything, just dumb. Somebody wants to waste their time doing that instead of reading the actual material, let them.
New Android phones are prominently displayed as having Android here in Japan. The latest AU commercial for one phone (the Sharp IS03) actually headlines Android and mentions the phone name only in passing. So yes, public awareness is probably pretty high.
"I honestly can't think of anything more useless than an OS that will not work if you don't have an internet connection."
Like when you're writing? Or playing some casual game? Or editing an image, or a drawing? Working on a spreadsheet or presentation? Programming or working with an analysis tool like R, Matlab or Maxima?
I like the net. I use it constantly. It's a great communication tool, information source and distractor. But to be honest, very little of my actual work needs the net. I could probably restrict access to one hour in the morning every day and still remain exactly as productive as I am now; more, quite possibly.
And sometimes - when you're abroad, when you're travelling - you simply don't have net access. A net-only computer becomes a useless paperweight. A net-only netbook or tablet would interrupt me every day on my commute when the train loses connection in tunnels.
A OS that requires a net connection to work would be pretty damn useless to me.
"Yeah, but what are you DOING with all that data? There is no display in the world that will show it, and it's even insane for any realistic printing. The only reason for resolution like that is archival."
It's the equivalent of a RAW file. I'd not use it directly of course - as you say, even a normal 15Mp digital camera file is basically more than any of us normally needs. But just as with a RAW file, you want to start with the full-size file to do your postprocessing if you can.
You may well want to crop just part of a scene, for instance, and later - months later - return to the image with fresh eyes and decide that a completely different crop really works better. Once you start doing non-trivial amounts of cropping, resolution starts to drop quickly. I did a set of A1-sized prints for a demonstration (they were basically just backdrops so they weren't critical). I took those with a digital camera but regretted it; I really missed the extra cropping headroom a larger image would have given me.
You may also want to change the overall brightness, or decide that a different color balance really is better than the one you selected initially. And just as you have noise in a RAW file, you have noise - grain - in the scan that you may want to (or not) reduce or deal with in one form or another. If nothing else, you generally want to smooth the blue channel in color images to get rid of the speckled appearance of skies in film scans. Such processing is always best done at the highest real resolution and with all the bit depth you have to avoid aliasing artefacts.
And of course, with three full color channels (yes, the film has three full channels and the scanner really scans them) it'll still take three times the space of a RAW file with equivalent resolution, so reducing the size is only going to help so much.
This is getting rather far from the original subject, though; I should not have formulated my objection to this file format in this manner.
"No, you do not need 16bpc, 20000px^2 images on a goddamn web page."
No I don't. But for a web page, I don't need anything but Jpeg in the first place.
There's a rule of thumb that for a new technology to supplant an old, it needs to be 3 times "better", for whatever value of "better" you care about. Improvement needs to be dramatic or it won't have any uptake. The only thing this format seems to do is fail a bit more gracefully for really low-quality images; that's not "3x better", which again raises the question of what the point is of releasing this new format.
I brought up 16 bit encoding as one area where the format could have made drastic improvement and differentiated it enough from Jpeg, but in retrospect I should have formulated this in a completely different manner. And not post it right after coming home from dinner and beers.
"You'd also likely be best served by simply scanning at a lower resolution. What exactly are you doing that you need such giant images, since you're not archiving them (you have the negatives for that)? "
The practical resolution from a V700 is about 2500dpi. Which, with 6x7 format color images (56x70mm) is about 120mb. If I save half the frames on a roll (that would be typical), I'll end up with half a gigabyte of data. No "going to 11" here.
And yes, I do have the negatives, but it's rather a pain to have to go back to that if I want to do a second edit. The point of what I want to do is to store one version in a good enough format that I don't have to go back to the negative just because I want to do a different crop, change the color balance or something.
What do I do now? I throw away the full-size images. Which mean I can't easily revisit an image later to try out a different edit.
"RAW from a Bayer sensor is certainly not 10-15 times the size of max quality JPEG"
OK, You're right. I get about a 7-8x difference: a PEF file of a busy scene (lots of leaves and stuff) from my Pentax is 13Mb while the Jpeg from UFRaw is 1.7mB, or about 7.5x. Quieter scenes (with lots of sky and stuff) have less of a difference. So 10-15x was rather an exaggeration. It's not about the same size either, on the other hand.
Which is beside the point in either case, as scanned images still are and remains too large to save at the full practical resolution for me without some form of high quality lossy encoding.
"So in summary, you seem to be someone who has a fetish for insanely high resolution digital images but wants to do horrible things like lossy compression to them, "
Lossy compression is surprisingly non-horrible when done well. At an earlier project we had to move a lot of images really fast through a set of network links (a distributed visual processing system). We had problems with network saturation so we tested various compression methods. And surprisingly, high-quality Jpeg was just fine. Even after multiple passes of compression and processing there was no visible hint of jpeg artifacts.
I willingly admit film is a weird, small corner case. But I wouldn't be the only one to benefit from a "jpeg but better" format. Anybody who's made a panorama with Hugin or similar knows just how large the final image becomes. And it would be nice to be able to save that full-size image for later cropping or re-editing, but a TIFF file of hundreds of megabytes isn't something you can easily store. And if you do multiple panoramas you simply can't. With a 16-bit version you could change the color balance and the brightness afterwards if you wanted, without visible aliasing.
Again, this is a side track. The main point is, they are introducing a new file format without any obvious improvements on the Jpeg format it's supposed to replace. And the restriction of 16k pixels to a side means it will be useless for some uses - again, you can easily get a panorama that size if you go all enthusiastic with Hugin for instance.
"For the majority of uses, high quality images should be stored as RAW. They're similar in size to high quality JPEG,"
No, they are not. It's a factor ten or fifteen between any lossless compression and high-quality (say, 95 or 97) Jpeg. For a digicam with ten megabyte RAW files it doesn't matter. But I shoot MF film, and one scanned color frame (three full color channels; no Bayer interpolation) is about 350Mb. A BW frame is about 100Mb. Which is too much to save. Those people using MF digital backs have less severe issues due to Bayer interpolation, but even there you're looking at upwards of 50-70Mb per frame.
Now, I do have the negative, so I can always rescan if I need the full image again. But usually I just need the full-resolution image, in a good enough format that I can reedit it for a particular use. Having a pre-processed version available in a lossy format at high quality is plenty good enough for that. If, that is, the format does save higher bit depths, and doesn't restrict the resolution the way this one does.
"If you do want some compression, choose from among the host of image file formats. If it's for your own archiving there are a huge number of choices."
I'm all ears. Seriously. What alternative can I use that does lossy encoding with good control over the quality, like Jpeg, but that lets me use 16 bits per channel rather than 8? And at minimum it has to have been implemented both for reading and writing by ImageMagick or some other similar open tool (so it doesn't disappear when some single vendor goes belly-up, or gets sued out of existence) so that I can at least convert to and from a lossless format for processing. If you have something I would really, seriously love to know about it.
Jpeg is mostly fine, even for high-quality real-world images if you keep to a high quality setting (95 or more). No visible artefacts, but a factor ten compression compared to good lossless formats. If you, say, want to store a copy of your film scans or other large images, high-quality lossy encoders can save you an enormous amount of space.
But there's one single problem with using Jpeg for that today: it doesn't support more than 8 bit data per channel. That makes it kind of suck as a lossy format for high-quality images. Any subtle tone shifts - a clear sky, for instance, or a near-white background - and you have visible banding. We have a possible contender - Jpeg2000 - that's been mostly rejected in the market due to the heavy thicket of patents surrounding it.
So what does Google do? They propose another image format without higher bit-depth support. And just to make sure that it's really useless for high-quality image applications, they restrict the image size to no more than 16383 pixels on a side. Anybody who has even played with panoramas with Hugin knows that you can easily get images larger than that.
So really, what is the freaking point? It doesn't do anything Jpeg isn't already doing well enough.
Slashdot Mirror
It being Nature group is no guarantee of success or high impact. And we have no idea if they are in it for the long haul or if they'll bail in a few years if the uptake is low. I'd just wait a couple of years and see what happens to it before submitting a paper there. Meanwhile, PLoS has a good impact factor, large readership and doesn't have a limit on the number of accepted papers so that's a better option for now.
External harddrive. Second harddrive in different location and/or a flickr account so you don't have it only at home.
Now, seriously? I have photography as a hobby too, and in the very best case about one in ten shots are actually worth saving, and that proportion drops the more shots you take. Most pictures end up being crap - that's not an indictment, just a fact. Most shots Adams or Bresson took ended up as crap too. You have perhaps two hundred pictures, tops, that you or anybody else would care about.
Chances are you're never ever going to look at most of those thousands of shots ever again, and your kids will simply throw it all away unseen in the far future when they're cleaning out your belongings. Problem is, if you haven't edited the collection and thrown away all the thousands of duds, they will end up throwing away those good, important images along with the rest.
So, best approach: go through and delete all the crap, all the duplicates, all the technical duds. Then delete all the ho-hum images. Aim for, say, two hundred images to save, or better, fifty. No matter how eventful a vacation, you don't have more than fifty great places or events to record. And the more pictures you keep, the more you dilute the impact of the good images.
Anything that needs me to manually adjust the glasses is a no-go for me. Normal progressive lenses work just fine already, so I don't really see the point of making it complicated and intrusive. If I want to improve my eyesight I'd much rather get surgery for my shortsightedness and astigmatism, then use progressives with uncorrected upper area. Either way I don't have to mess with my glasses just because I want to change my viewing distance.
Hub-and-spoke designs can be really frustrating. The problem is utilization, though: the town center is by far the most common destination so it's the one that can support most traffic. Even if most people want to go from one peripheral point to another, they all want to go between different points. A normal bus line would be too little used to be realistic.
You can't really adjust capacity all that much by spacing out the vehicles; people will only want to wait for so long to take a bus, and don't like having to plan their day around a bus time table. And you can't adjust the vehicle capacity too much either. A ten-passenger bus costs almost as much to run as a 50 passenger one. Driverless vehicles would be able to change that situation, but we're not there yet.
Armchair discussions on language efficiency, convenience and so on are really fun. Plenty of people argue that Japanese would be better without kanji. But the actual results is what matters.
We're collectively lazy, and our languages tend to become reasonably efficient over time. You can see how word length roughly corresponds to usage frequencies, how phoneme distribution and spelling tends to strike an efficient balance between brevity and redundancy so errors are minimized. None of this is consciously decided by anyone; instead people tend to gravitate over time toward usage patterns that minimize errors, redundancy and effort.
Japanese has, as you say, not just one but two syllabaries in addition to kanji, and has had them readily available for more than a millennium. A century or so ago there was an official drive to reduce the usage of kanji, with an eye towards abolishing them altogether. There has been lots and lots of time, and lots and lots of opportunity to move away from kanji in other words. So what happens?
People are using more kanji today, not less. Not from a sense of tradition, or from some official decree, but because they like to do so. They add expressiveness, and are another resource for punsters and jokers. The culture and education department recently acknowledged this trend and added another two hundred characters to the list that people are required to study at school.
Part of the reason use is increasing is because of electronic communication. Many of the drawbacks of kanji - it's difficult to remember how to write a rarely used character, for instance, or remember a rare pronunciation - are mitigated with cellphones and computers. It's much, much easier to recognize a character than two write it by hand, and with modern input methods that's all you need. And an unfamiliar word is easy to look up directly on the same device you're reading it on.
People are voting with their feet, as it were, and moving towards more kanji, not less.
You're way behind, Mr. Mopps. Behold: http://linnnk.com/awesome-minecraft-costume
How about using rsync? That should cut down a lot on the copy time. I haven't tried it myself, mind you; my world is still small enough that a straight copy takes very little time.
Well, the passage is an aside set in the middle of a sentence about farmers. Looks pretty clear to me, but since it's not clear to others I will try to be a bit more mindful about such things in the future. It kind of sucks when your argument misses the point because your text ends up hard to understand.
Yomiuri Shinbun just ran a series of articles looking at the current state of agriculture, using official figures as a base. Here is one relevant piece in English (warning: articles disappear offline after a few days): http://www.yomiuri.co.jp/dy/business/T101130005037.htm
Here's the relevant part:
You may get different statistics if you count farmland owners rather than farmers. A lot of the uncultivated land is plots that have been inherited by people who live and work in the city. They live too far away to farm the land on one hand, and they don't want to sell the family land on the other, so it ends up sitting unused.
Nope. There's plenty of farmland around in Japan; a fair amount of high-quality farmland lies fallow, in fact, from a lack of interest in using it. The reasons for the interest in factory farms (the indoor farming you've seen) and robotics are somewhat different. It's potentially much more effective and with higher-quality yield than open-field farming, where each plant gets the optimal amount of light and nutrients, no pests and no pesticides, and the same results no matter what the season or weather. And you can have small-scale farms that are still efficient, so a large restaurant could actually grow some of their own produce to the exact standards they want, all year round.
It would take fewer people to work such a farm too, and even more so with robots. The major reason for fallow land in Japan is the lack of people willing to be farmers in a distant rural area - the average age is already above retirement age - so this would be a way both to reduce the number of people needed, and move the industry to where people want to live and work.
This is what I meant about overselling this idea. A GPU is not an alternative to a cluster. The practical speedup you get is 2-3 times a normal dual-core cpu (a bit more for specific problems, less for others). A GPU is in other words very roughly equal to adding another 2-core cpu to your system, which makes the cost-benefit tradeoff clearer: cheaper node, but longer development time. If you need a cluster for your problem you're going to be using MPI no matter what*; one or two GPUs will make no material difference.
What I am talking about is clusters equipped with GPUs at each computing node. At which point you need to choose between spending the extra time developing for those GPUs; spending extra time on longer simulations without them; or spending extra money to get access to more computing nodes. And what I see (personal experience only, so not very reliable) from the clusters I have access to is that most researchers skip the GPUs in favour of the other two solutions today.
* Actually, some clusters can handle threaded applications across nodes, and that can be an alternative. But many massively parallel problems are really better expressed with something like MPI anyhow, and MPI is close to a true distributed processing standard. Use MPI on linux and your code is fairly easy to port between machines.
We're talking supercomputers, not pc:s. A lot of the software used there really is written for one single task or one single project. Once the project is over and the original user is done with it, it's never used again. While you may want to do something similar at a later date your new task is different enough - and the new machine you use is different enough - that you'll need to redesign the application. And if you were not part of the previous team, understanding and adapting their code (never mind finding the code in the first place) to your machine may well be more work than simply developing your own software from the start.
So the relevant measure is the total time spent during the lifetime of that project, and that means development time really does become a major factor to take into consideration.
GPUs are indeed an inexpensive way to boost speed in some cases. But they have been rather oversold; while some specific types of problems benefit a lot from them, many problems do not. If you need to frequently share data with other computing nodes (neural network simulations come to mind), then the communications latency between card and main node eats up much of the speed increase. And as much of the software you run on this kind of system is customized or one-off stuff, the added development time in using GPUs is a real factor in determining the relative value. If you gain two weeks of simulation time but spend an extra month on the programming, you're losing time, not gaining it.
Think about it this way: GPU's are really the same thing as specialized vector processors, long used in supercomputing. And they have fallen in and out of favour over the years depending on the kind of problem you try to solve, the relative speed boost, cost and difficulty in using them. The GPU resource at the computing center is used much less than the general clusters themselves, indicating most users do not find it worth the extra time and trouble to use.
It is a good idea, but it's not some magic bullet.
Yes, we are. Our vision system is pretty successful when you look at how we actually use it in the real world. We don't actually need to know the precise distance to things; what we want to know is rather direction and time to impact and similar and we're really, really good at that (look up tau-margin estimation for instance). Though note that with a human-level vision system you would still need a lot of those sensors you talk about. Our vision system absolutely depends on proprioception to figure out where we are in the world and compensate for our own movements; we need separate dead-reckoning systems and (again) a lot of experience to be even somewhat correct about our movements over large distances and so on.
But I wrote this in reply to a poster that seemed to believe we humans are actually better than Kinect at the specific vision tasks it's built to do. Too many people seem to believe that the mammalian vision system is inherently great, at whatever tasks we imagine, and that if we could only make something like it our machine vision problems would be solved. That is simply not the case.
The "good ol' brain" does a fairly crappy job, actually. 3D vision systems like these tend to perform quite a bit better than we do. And we only do as well as we do because we can use a lot of indirect clues based on our long experience with a 3D-world - we know how big stuff normally is, for instance, so we can judge distance from size. Mess up those clues and we completely lose it.
And even with good clues we don't actually measure distance well. Have somebody place items on a parking lot or some place like that, then try to guess the distances. Not going to be very accurate. Try to estimate distance vertically rather than horizontally and you'll do even worse; you have fewer clues and less experience to fall back on.
Well, it is a completely different drink when you down the vodka neat then drink the orange juice as a chaser. The difference may be lost on many four-year olds I suspect, though.
Wbhy would he be punished? Memorizing the answer order isn't unethical or cheating or anything, just dumb. Somebody wants to waste their time doing that instead of reading the actual material, let them.
New Android phones are prominently displayed as having Android here in Japan. The latest AU commercial for one phone (the Sharp IS03) actually headlines Android and mentions the phone name only in passing. So yes, public awareness is probably pretty high.
"Area Data Center Actually Located in Data Center Facility; IT Experts Confused, Baffled"
"I honestly can't think of anything more useless than an OS that will not work if you don't have an internet connection."
Like when you're writing? Or playing some casual game? Or editing an image, or a drawing? Working on a spreadsheet or presentation? Programming or working with an analysis tool like R, Matlab or Maxima?
I like the net. I use it constantly. It's a great communication tool, information source and distractor. But to be honest, very little of my actual work needs the net. I could probably restrict access to one hour in the morning every day and still remain exactly as productive as I am now; more, quite possibly.
And sometimes - when you're abroad, when you're travelling - you simply don't have net access. A net-only computer becomes a useless paperweight. A net-only netbook or tablet would interrupt me every day on my commute when the train loses connection in tunnels.
A OS that requires a net connection to work would be pretty damn useless to me.
"Yeah, but what are you DOING with all that data? There is no display in the world that will show it, and it's even insane for any realistic printing. The only reason for resolution like that is archival."
It's the equivalent of a RAW file. I'd not use it directly of course - as you say, even a normal 15Mp digital camera file is basically more than any of us normally needs. But just as with a RAW file, you want to start with the full-size file to do your postprocessing if you can.
You may well want to crop just part of a scene, for instance, and later - months later - return to the image with fresh eyes and decide that a completely different crop really works better. Once you start doing non-trivial amounts of cropping, resolution starts to drop quickly. I did a set of A1-sized prints for a demonstration (they were basically just backdrops so they weren't critical). I took those with a digital camera but regretted it; I really missed the extra cropping headroom a larger image would have given me.
You may also want to change the overall brightness, or decide that a different color balance really is better than the one you selected initially. And just as you have noise in a RAW file, you have noise - grain - in the scan that you may want to (or not) reduce or deal with in one form or another. If nothing else, you generally want to smooth the blue channel in color images to get rid of the speckled appearance of skies in film scans. Such processing is always best done at the highest real resolution and with all the bit depth you have to avoid aliasing artefacts.
And of course, with three full color channels (yes, the film has three full channels and the scanner really scans them) it'll still take three times the space of a RAW file with equivalent resolution, so reducing the size is only going to help so much.
This is getting rather far from the original subject, though; I should not have formulated my objection to this file format in this manner.
"No, you do not need 16bpc, 20000px^2 images on a goddamn web page."
No I don't. But for a web page, I don't need anything but Jpeg in the first place.
There's a rule of thumb that for a new technology to supplant an old, it needs to be 3 times "better", for whatever value of "better" you care about. Improvement needs to be dramatic or it won't have any uptake. The only thing this format seems to do is fail a bit more gracefully for really low-quality images; that's not "3x better", which again raises the question of what the point is of releasing this new format.
I brought up 16 bit encoding as one area where the format could have made drastic improvement and differentiated it enough from Jpeg, but in retrospect I should have formulated this in a completely different manner. And not post it right after coming home from dinner and beers.
"You'd also likely be best served by simply scanning at a lower resolution. What exactly are you doing that you need such giant images, since you're not archiving them (you have the negatives for that)? "
The practical resolution from a V700 is about 2500dpi. Which, with 6x7 format color images (56x70mm) is about 120mb. If I save half the frames on a roll (that would be typical), I'll end up with half a gigabyte of data. No "going to 11" here.
And yes, I do have the negatives, but it's rather a pain to have to go back to that if I want to do a second edit. The point of what I want to do is to store one version in a good enough format that I don't have to go back to the negative just because I want to do a different crop, change the color balance or something.
What do I do now? I throw away the full-size images. Which mean I can't easily revisit an image later to try out a different edit.
"RAW from a Bayer sensor is certainly not 10-15 times the size of max quality JPEG"
OK, You're right. I get about a 7-8x difference: a PEF file of a busy scene (lots of leaves and stuff) from my Pentax is 13Mb while the Jpeg from UFRaw is 1.7mB, or about 7.5x. Quieter scenes (with lots of sky and stuff) have less of a difference. So 10-15x was rather an exaggeration. It's not about the same size either, on the other hand.
Which is beside the point in either case, as scanned images still are and remains too large to save at the full practical resolution for me without some form of high quality lossy encoding.
"So in summary, you seem to be someone who has a fetish for insanely high resolution digital images but wants to do horrible things like lossy compression to them, "
Lossy compression is surprisingly non-horrible when done well. At an earlier project we had to move a lot of images really fast through a set of network links (a distributed visual processing system). We had problems with network saturation so we tested various compression methods. And surprisingly, high-quality Jpeg was just fine. Even after multiple passes of compression and processing there was no visible hint of jpeg artifacts.
I willingly admit film is a weird, small corner case. But I wouldn't be the only one to benefit from a "jpeg but better" format. Anybody who's made a panorama with Hugin or similar knows just how large the final image becomes. And it would be nice to be able to save that full-size image for later cropping or re-editing, but a TIFF file of hundreds of megabytes isn't something you can easily store. And if you do multiple panoramas you simply can't. With a 16-bit version you could change the color balance and the brightness afterwards if you wanted, without visible aliasing.
Again, this is a side track. The main point is, they are introducing a new file format without any obvious improvements on the Jpeg format it's supposed to replace. And the restriction of 16k pixels to a side means it will be useless for some uses - again, you can easily get a panorama that size if you go all enthusiastic with Hugin for instance.
"For the majority of uses, high quality images should be stored as RAW. They're similar in size to high quality JPEG,"
No, they are not. It's a factor ten or fifteen between any lossless compression and high-quality (say, 95 or 97) Jpeg. For a digicam with ten megabyte RAW files it doesn't matter. But I shoot MF film, and one scanned color frame (three full color channels; no Bayer interpolation) is about 350Mb. A BW frame is about 100Mb. Which is too much to save. Those people using MF digital backs have less severe issues due to Bayer interpolation, but even there you're looking at upwards of 50-70Mb per frame.
Now, I do have the negative, so I can always rescan if I need the full image again. But usually I just need the full-resolution image, in a good enough format that I can reedit it for a particular use. Having a pre-processed version available in a lossy format at high quality is plenty good enough for that. If, that is, the format does save higher bit depths, and doesn't restrict the resolution the way this one does.
"If you do want some compression, choose from among the host of image file formats. If it's for your own archiving there are a huge number of choices."
I'm all ears. Seriously. What alternative can I use that does lossy encoding with good control over the quality, like Jpeg, but that lets me use 16 bits per channel rather than 8? And at minimum it has to have been implemented both for reading and writing by ImageMagick or some other similar open tool (so it doesn't disappear when some single vendor goes belly-up, or gets sued out of existence) so that I can at least convert to and from a lossless format for processing. If you have something I would really, seriously love to know about it.
Jpeg is mostly fine, even for high-quality real-world images if you keep to a high quality setting (95 or more). No visible artefacts, but a factor ten compression compared to good lossless formats. If you, say, want to store a copy of your film scans or other large images, high-quality lossy encoders can save you an enormous amount of space.
But there's one single problem with using Jpeg for that today: it doesn't support more than 8 bit data per channel. That makes it kind of suck as a lossy format for high-quality images. Any subtle tone shifts - a clear sky, for instance, or a near-white background - and you have visible banding. We have a possible contender - Jpeg2000 - that's been mostly rejected in the market due to the heavy thicket of patents surrounding it.
So what does Google do? They propose another image format without higher bit-depth support. And just to make sure that it's really useless for high-quality image applications, they restrict the image size to no more than 16383 pixels on a side. Anybody who has even played with panoramas with Hugin knows that you can easily get images larger than that.
So really, what is the freaking point? It doesn't do anything Jpeg isn't already doing well enough.