sdai Posted October 24, 2006 Share #121 Posted October 24, 2006 Advertisement (gone after registration) I really appreciate your comments, Andrej ... and thanks for your time and effort putting into this. Contrary to many internet friends' belief, I'm not a bit into these computer stuff ... but, with the help of my kids, I was able to run dcraw, EXIF tool and imagemagick identify against the leaked dng file widely spread on the web ... I would say, your observation certainly isn't unfounded as some posters tried to imply and raised some more interesting questions in this regard. Although we could do some sort of investigation and proofing on our own, like Scott has suggested, I surely want someone in the know could give an "official" word and convincing explanation. Link to post Share on other sites More sharing options...
Advertisement Posted October 24, 2006 Posted October 24, 2006 Hi sdai, Take a look here M8-why 10MB-vs-DMR 20MB . I'm sure you'll find what you were looking for!
plasticman Posted October 24, 2006 Share #122 Posted October 24, 2006 I also appreciate this open debate - it certainly appears that there's an element on the forum of hoping that the whole discussion would simply disappear... In any case, it looks like i'll probably have to pay hard currency tomorrow in order to guarantee my place in the preorder queue, and so far i haven't seen any real evidence to contradict the fact that it APPEARS that Leica may plausibly be 'nuancing' the facts in their marketing material for the camera. Notwithstanding that the photographs i take would, in all probability, look equally poorly taken if i was using a pro-level camera capturing 16bits or a cellphone, i would still like to know what i'm paying for. Only fair i think. Link to post Share on other sites More sharing options...
Kaj Jensen Posted October 24, 2006 Share #123 Posted October 24, 2006 Is the conclusion 8 bit? The dynamic range of the CCD mapped into 256 steps by a lookup-table. The nonlinear mapping might introduce large intensity steps in bright areas – does a crop in photoshop of a bright area show banding? Is banding visible after sharpening ( FocalBlade, nik-Sharpener-Pro or other)? Link to post Share on other sites More sharing options...
sean_reid Posted October 24, 2006 Share #124 Posted October 24, 2006 Andrej, your last post before this made it perfectly clear: the stored data bytes are the square roots of the as-shot bytes, with some bit shifting to preserve significant data as much as possible. The restored data is 15 bits, from squareing the stored bytes and shifting back two places. That's a sensible thing to do, and with care and good raw file development software most of us will probably never see any artifacts. As a benefit we are getting twice as many pictures in the RAW file buffer, faster review speed on the LCD and maybe higher continuous shooting speed. The Leica documents do seem to be misleading. Sean Reid is aware of the situation and will ask Leica what's going on sometime fairly soon. If there's anything to be said about it, I expect he will do so. Sorry you are getting ignored while other preach about what Leica must do, but Leica's just a company with good hard-working engineers, the M8 is just a camera (but apparently a damned good one), and netizens will be netizens. If you want to model things, could you see if dithering by adding low order bits in the stretched upper tones can help prevent banding in the skies and highlights? regards, and good luck with your raw file development package. Does it have a sexy name yet? scott To be specific, I'm aware of the questions about this and will be talking with Leica about to get information to clarify this aspect. I'll then included a section on this topic either in part 3 or as an addition to Part 2. Meanwhile, for those who are following the review series, take a look at the ISO 160 crops I just added - wow! Whatever the facts may be about the RAW files, the results sure look good. Cheers, S Link to post Share on other sites More sharing options...
plasticman Posted October 24, 2006 Share #125 Posted October 24, 2006 Meanwhile, for those who are following the review series, take a look at the ISO 160 crops I just added - wow! Whatever the facts may be about the RAW files, the results sure look good. Totally agree with that final statement anyway! Those crops are truly awesome. And i thought before that your comparisons with the 5D showed the M8 has the advantage at lower ISOs - the detail is simply better rendered and the colors seem more 'lifelike' right up to (and including) 640. All the same, there are a few small nagging uncertainties about the camera that've cropped-up over the last few weeks that are somewhat unsettling - and this bitdepth question is a typical example that makes buyers who are sinking $5K into the M8 a little nervy, to say the least. At last we have some decent images to look at - and that is helping to cement the decision - but the fact remains that Leica may not have been unimpeachably straightforward in their promotional material. Link to post Share on other sites More sharing options...
sdai Posted October 24, 2006 Share #126 Posted October 24, 2006 I pulled off the DNG files Roberto uploaded on to his site and found that the production firmware 1.06 still outputs 8-bit uncompressed files. Link to post Share on other sites More sharing options...
marknorton Posted October 24, 2006 Share #127 Posted October 24, 2006 Advertisement (gone after registration) Is there anyone here who wouldn't like to have the option to take the hit on write times, buffer and card capacities, shooting speed and battery life and write uncompressed 16 bit images instead? Using the y = Sqrt(4 * x) and x = (y^2)\4, I've written a program to see just what errors are being introduced and while the actual results will depend on what rounding and dithering they apply, the fact is that information is lost and there's no way to get it back. In effect, the discrepancies introduce a layer of noise which is only loosely correlated to the image data caused by word length truncation. The biggest percentage errors are in the black areas, the biggest absolute errors are in the bright scenes, as you would expect. It's a well known principle in Digital Audio that every time you truncate your word lengths, you loose information. Link to post Share on other sites More sharing options...
plasticman Posted October 24, 2006 Share #128 Posted October 24, 2006 EXACTLY Mark! Information is lost in that first stage - and what is even worse, any post-processing inevitably multiplies the consequential loss. Some people may consider this all academic. I don't. Link to post Share on other sites More sharing options...
Andrej Kolev Posted October 24, 2006 Share #129 Posted October 24, 2006 Exif tool is right, AA filter strength is 1.0, at first time I havent look to a right offset. It means strong anti-alias filter, but according to specs published DPReview "No anti-alias filter (low pass filter) ". Link to post Share on other sites More sharing options...
marknorton Posted October 24, 2006 Share #130 Posted October 24, 2006 Information is lost in that first stage - and what is even worse, any post-processing inevitably multiplies the consequential loss. Yes, see that blinking red light on the back of the M8? The meaning of that light is "Warning - I'm discarding 43% of your data and there's nothing you can do to get it back". What I cannot prove is that keeping the data will improve the final image quality noticeably. What I can say with 100% certainty is that it will not make it worse. Link to post Share on other sites More sharing options...
sdai Posted October 24, 2006 Share #131 Posted October 24, 2006 Hi, Andrej, you've just edited your post right before I hit the reply button ... yes, I was using EXIF tool to look at the file. I've also tried DCRAW my little one compiled for me but it wasn't able to yield more information. Link to post Share on other sites More sharing options...
EricC Posted October 24, 2006 Share #132 Posted October 24, 2006 Yes, see that blinking red light on the back of the M8? The meaning of that light is "Warning - I'm discarding 43% of your data and there's nothing you can do to get it back". What I cannot prove is that keeping the data will improve the final image quality noticeably. What I can say with 100% certainty is that it will not make it worse. Reading this thread i am starting to wish that i had not sold my Epson R-D1. Link to post Share on other sites More sharing options...
Jamie Roberts Posted October 24, 2006 Share #133 Posted October 24, 2006 Is there anyone here who wouldn't like to have the option to take the hit on write times, buffer and card capacities, shooting speed and battery life and write uncompressed 16 bit images instead?{snipped}. Or--better yet--is there absolutely anyone at all who would care if the files are different in size for different ISOs or shooting situations? I mean, who would really care that the files are all the same size!?!! That seems to be the design point, and it's just wrong, IMO. Lossless compression is fine and good and a great way to speed things up. At low ISOs, with little noise, you'd approach the current compression level anyway, and you wouldn't lose any information. Yes, the files sizes would vary, and you might get 10% fewer pics on an SD card. But for anyone buying an M8 the cost of SD cards is, well, neglible. Personally, at high ISOs, I couldn't care less if I need a larger SD card. Heck--I'm shooting the DMR anyway; I just pretend to myself it has the same file size as a 1ds2 and it doesn't bother me (ok, speed of review is sometimes a hit, but that could be changed in other ways). Yes, the DMR is slow. But the files it produces are wonderful. I hope the M8s are as good... Link to post Share on other sites More sharing options...
marknorton Posted October 24, 2006 Share #134 Posted October 24, 2006 Jamie, I agree. You can't accurately display the remaining picture count if you don't know how big they are going to be but I think most would accept an estimated figure based on the average of the pictures currently on the card or for the resolution selected. Such a compromise should not stop them doing it. The following three options would give us the best of all worlds: - Uncompressed 16 bit - Lossless 16 bit compression - 16->8 bit Lossy compression Lossless compression is pretty compute intensive and it comes back to my first post that "pk-zipping" each image might kill the battery and so too might writing 20Mb to the card. The M8 battery is 7 Wh, the DMR battery is nearly double at 13.2 Wh... Link to post Share on other sites More sharing options...
Andrej Kolev Posted October 24, 2006 Share #135 Posted October 24, 2006 Then we estimate loss here, we cannot think about ideal data coming from sensor, in reality sensor data contains noise, so thigs are not so simple. I think Leica has done some research before make a desicion to use nonlinear coding. This sceme is proposed by Adobe in DNG, so maybe they have convinced Leica to use it, or maybe Leica have made their own research. I think Leica will say our engeneers have found that for this sensor there is no noticible loss in precision. But it would be good to see some research about this problem. Link to post Share on other sites More sharing options...
Andrej Kolev Posted October 24, 2006 Share #136 Posted October 24, 2006 Some more thougths: in ccd's photon noise is estimated as a squre root of signal. Link to post Share on other sites More sharing options...
plasticman Posted October 24, 2006 Share #137 Posted October 24, 2006 Some more thougths: in ccd's photon noise is estimated as a squre root of signal. There's some incredible knowledge of the mechanics of image-capture here - and i actually think that Andrej may be quite right in assuming that the Leica engineers have tried to work towards the 'best compromise' for all-round use. But i absolutely second Mark's suggestions about user-defined options for compression ratios. Incidentally, i left the office rather earlier than usual to swing by my dealer - luckily they weren't wanting any deposit just yet, because they still don't know how many cameras they're gonna get!! :-) Link to post Share on other sites More sharing options...
sdai Posted October 24, 2006 Share #138 Posted October 24, 2006 This sceme is proposed by Adobe in DNG, so maybe they have convinced Leica to use it, or maybe Leica have made their own research. I can hardly believe that Adobe could have such an influence in this case but I agree, it may be the best compromise Leica's engineers could reach due to some limitations in the design. Link to post Share on other sites More sharing options...
blado Posted October 25, 2006 Share #139 Posted October 25, 2006 Exiftool output from a Canon cr2 file shows the following: Bits Per Sample : 8 8 8 This makes me think of an 'r g b' type of information thing. Link to post Share on other sites More sharing options...
ho_co Posted October 25, 2006 Share #140 Posted October 25, 2006 I can hardly believe that Adobe could have such an influence in this case but I agree, it may be the best compromise Leica's engineers could reach due to some limitations in the design. Since Leica worked directly with Phase One, I guess you're suggesting that Adobe influrenced them, and they Leica? As I said above, I respect the amount of understanding you guys have of the topic; but whatever the truth of the matter, Leica certainly seems to have turned out a surprisingly good image producer. For that reason I would hesitate to call it a 'compromise ... due to some limitations in the design.' Keep this line of inquiry going. S Reid will have some kind of answer soon from Solms, and we'll all be the beneficiaries of having read this thread! Technical speculations are fun of themselves, and seeing them explained and defended gives the rest of us a lot more understanding. --HC Link to post Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.