Slashdot Mirror

Tineye's image similarity is a lot smarter than Google's.

Smarter how, exacly?

For every image that I tried in the past year or so, Google found more matches than Tineye.
More often than not, Tineye would find zero matches while Google presented several pages.

Let's try it with an image from your site:
Google
Tineye (will expire in 72 hours)

OK, just for fun, let's take an image from the Tineye web site:
Google
Tineye (will expire in 72 hours)
Outright embarrassing.

Or a different one:
Google
Tineye
Ouch!

Plus, Google supports filtering results by image size and/or by time.
Not to mention that, no doubt due to the integration with their traditional search, they find related keywords that allow you to dig deeper (for example, find such images on non-indexed sites).

On the other hand, Google's results are arranged then Tineye's (I often have to click the "more sizes" and "visually similar" links to get to results that don't display on the main results page) and Tineye has more sorting options, but those are really minor points.

Do you know that Tineye once had a forum on its site? They took it down after it started getting filled with posts comparing it to Google Image Search and reaching the same conclusion -- that Tineye is no longer relevant.

The price isn't that bad, it's instructive, I think, to realize that using memory cards as one-time disposable resources is arguably half the cost of storing photographs on slide film used to be. For that price, one could reasonably send a memory card home and keep one with you, which seems a completely sensible backup policy. And, of course, you *can* reuse the cards.

I feel your analogy is poor.

If instead of stealing a digital copy of one of my Winter scenes of Mono Lake, you make a copy and then hand it to the Mono Lake Committee, then the Mono Lake Committee, when it publishes tens of thousands of copies of its calendar, hands you $250 instead of handing me $250.

If you grab an image and use it on a web site, the numbers are smaller, but still contribute to my budget for things like food and camera equipment. I know several photographers who have entirely pulled their work off the web because of image usage without permission.

There is no question that copyright is a social convention, one that people have "agreed" has some putative benefit to society. And I suspect you and I might be able to find some sympathy for each others opinion in how ridiculously exaggerated copyright law has become. But I don't think copyright law, in and of itself, is bad, and given what I do for a living, it's pretty unlikely you're going to convince me that I'm wrong.

With P1C1, I could load up a directory of a few hundred, let it chug some thumbnails while I stepped out for coffee, and really speed through the selection process, it probably saved me days of time for that shoot alone. A faster computer would have helped some, but not nearly as much as a better workflow.

Now, that having been said, I started using P1C1 for a lot of my conversions as well, because I end up making better conversions with it than I do with ACR. I don't think ACR is actually technically inferior, and in some important ways it's superior (I miss chromatic aberration reduction in P1C1, and will occasionally do conversions in ACR just for that when it's important), but for some reason I have an easier time dialing in a good conversion with P1C1, even after selection. It sounds to me as if Aperture might not be at this point, but I could imagine a later version filling very much the need I originally bought P1C1 for.

<geeksize>He should try it, then, I'll look forward to seeing one of his immages in the Smithsonian. </geeksize>

What's tragic about what Tom is suggesting is that you CAN'T reshoot a wedding.

Yeah, there's a smaller but real extent to which thats true in my work as well. Many of my best images are quite dependent on particular whether conditions. As an extreme example, This image gets its effectiveness from light sneaking through a hole in a blanket of clouds in the direction of sunrise, at sunrise, a condition which lasted around 30 seconds and which I haven't seen again there--and clear skies don't produce anything nearly as nice.

Not that it's nearly the same situation in general, I don't have to deal with clients emotionally invested in a single moment, (they never see the inevitable images that don't make the cut in my biz), it's just quite a different thing.

And yes, it is tragic.

My first test printing comparable images from drum-scanned film and directly from the 1Ds showed far better results at 24"x16" prints from the 1Ds--and then I realized I'd been shooting the 1Ds at ISO 400. My jaw dropped.

It turns out that the size of the scan stops being meaningful because of film grain, and that the amount of film grain turns out to make more of a difference in producing large prints than you'd expect from the "line pairs per inch" measurements. The grain just kills you trying to make enlargements, the cleanliness of the 1Ds image results in larger prints that come closer to very high-qulilty medium-format prints than 35mm.

I did a quick "review of the 1Ds for film photographers", it still needs some work, but you're welcome to read it.

The price tag is large for the 1Ds. For me, that's not an issue, it pays for itself in reduced film and processing costs in a year or so--and I get better results--and I get better flexiblity.

To my eye, sensors have well surpassed color film. Black and white is going to be closer, but ... I don't think you should rely (if you are, I don't mean to put words in your mouth) on "megapixels" from film scans as being comparable to "megapixels" from digital sensors directly. Do the experiment, you might be surprised.

Long answer: Actually, most fine art nature photography is done with computers in the loop now. The problem really begins with the non-linearity of the chemical print-making process. Slide film captures color and vibrancy from scenes far better than print film, unfortunately, traditional chemical methods for making prints result in very increased contrast.

Now some people do a lot of post-processing in the darkroom to fix this. For anything but strict documentary photoraphy this has always been the case, I urge you to read some of the more technical books by Ansel Adams as an example.

However, this particular problem can be solved with technology, and it's pretty damn cool technology, too. Scan the slide with a drum scanner (a device which rotates the slide with a single light source and a single PMT reading the slide, it takes forever but you can pull enormous info out of a slide. Create the final print using the same chemistry (controlled temperature, reagents, etc.) and paper but instead of optically enlarging through the print, use computer controlled lasers. (I am not making this up.) The effect of the laser on the paper can be calibrated, and so reproducable prints with more normal contrast can be made. I'll leave the math as an exercise to the reader.

Those Ansel Adams books will also describe a lot of dodging and burning. I do some of that. I don't move mountains or cigarette butts, because that's my ethic of nature photography, but I do believe that some local contrast enhancement such as that done by dodging and burning can produce prints which are subjectively more accurate to "what I saw" than the film process otherwise allows. More detail could follow here, but I think you get the idea.

This technology also applies to 4-color printing. I do my own color separations for my notecards. I give the printing house digital represntations of the C, M, Y and K plates they should create for those cards, and they send me the results. It's a wonderful thing.

The computer can also be used to correct for deficiencies of technology. For example, even the best lenses, with good enough sensors or scanning, can show chromatic aberration. This is an artifact where red light (say) is imaged onto the film or sensor differently than green or blue. Oddly enough, computers can help fix this distortion, although it's quite minor in prints less than about 24x16".

Since I do large prints (24x16) from film, but use digital printing already (I get high-resolution 100MB drum scans made from my slides), the change in workflow would be easy.

The math works out like this for me. The digital Rebel won't take pictures good enough to blow up to that 24x16 as well as film, but the 1Ds will. The 1Ds is about $8K, but it'd save me about $2K/year in film, developing costs, and drum scans. I'm already a Canon user so there'd be no cost for switching systems.

I'm betting that the 1Ds or equiv. will be more than $2K cheaper by the Summer of next year, so I'm still using film. But you can see where, depennding on how you think about the analysis, it might be the time.

From a resolution point of view only, there's little point to the 1Ds unless you're making prints above 10x15 or so.

Canon's CMOS-based sensors, which will be used in the D1s, have proven excellent color stability and tonality when used correctly as shown by folks like Michael Reichmann of the Luminous Landscape. Previous digital-SLRs that used those sensors (D30, D60) demonstrated excellent low-light performance and had smaller than "full-frame" sensors, the size of the individual pixels on the D1s won't be very dissimilar IIRC to those on the Canon D60.

Larger than 3-4 megapixel resolution does matter to me, but only because I want to make 24x16 prints. If you're happy with 8x10s, there's nothing wrong with 3-4 megapixels in and of itself (although not all 3MP cameras are created equal by any means.) Still, for regular prints there should be no reduced quality at all with proper data handling.

I do landscacpe photography, 8fps is overkill.

who's ever heard of a professional photographer printing a digital image in large format?

Moving images through a digital stage is already standard procedure for many fine art photographers who do image capture on film, folks like the late Galen Rowell already use this process (a workflow that was, interestingly enough, improved a lot with the work of former Mac Ghod Bill Atkinson. (Interestingly enough, these processes end up again back on silver nitrate paper, but I digress.)

Starting off with digital images would actually remove layers of "stuff" happening to the image reducing quality--so long as the orignial image is detailed enough (in spacial resolution, in contrast range, and in color resolution.) Existing sensors can achieve this, the missing link really was resolution.

The new Canon D1s (not to mention the Sigma SD9, the Kodak 14MP SLR, and the Kodak 16MP digital back for the to-be-announced-in-the-next-day Hasselblad H1) are going to take serious bites into the serious film photography market.