gabrielma Posted April 17, 2007 Share #161 Posted April 17, 2007 Advertisement (gone after registration) Oh, and also ask Leica to increase the area of coverage that the framelines show. Right now they are off by a good 20%, which is ridiculous, even for a rnagefinder. I'd like to point out that this may be a case of some shooters that may be experiencing parallax correction error. I know my M8 is off, after comparing it with two other bodies. My shots get chopped off at two edges: bottom and right, if viewed "Landscape", specially when focusing close. This can get very confused with "frameline coverage". I encourage those of you who have have been frustrated often by "chopped off" shots "at the edge" to investigate this on your own and align your shots on the edge of each frameline (I suggest the 50mm or 90mm framelines) and see where the image gets cut off as a rule. This is particularly annoying when you are someone who likes to compose through the viewfinder and not crop shots after the fact. I don't have to go through this with my film bodies. Link to post Share on other sites More sharing options...
Advertisement Posted April 17, 2007 Posted April 17, 2007 Hi gabrielma, Take a look here Official Response from Leica on Laundry List. I'm sure you'll find what you were looking for!
SoBeachKid Posted April 18, 2007 Share #162 Posted April 18, 2007 Thanks Guy ! Have you guys heard anything on this issue that was described there ? http://www.leica-camera-user.com/digital-forum/16122-m8-display-intensity-periodic-variations-when.html Link to post Share on other sites More sharing options...
Guest guy_mancuso Posted April 18, 2007 Share #163 Posted April 18, 2007 Just sent that in, this is new one is this on 1.102 Link to post Share on other sites More sharing options...
SoBeachKid Posted April 18, 2007 Share #164 Posted April 18, 2007 Thanks Guy. I have noticed this since the beginning (1.09 for me). I still see the issue with 1.102. Link to post Share on other sites More sharing options...
thrid Posted April 23, 2007 Share #165 Posted April 23, 2007 This is the graphic presented in the article of the LFI magazine. The origin is at the top left. The OX axis (horizontal) represents the intensity of the light, from shadows (at the left) to highlights (at the right). The OY axis (vertical) represents the number of tonal variants, from 0 (at the top of the graph) to infinite (at the bottom). To sum-up, the M8's DNGs provide more tonal resolution than a linear 12-bit file only in the deep shadows; more tonal resolution than the typical linear 8-bit files (JPG and TIFF) in the midtones; and even less information than linear 8-bit files in the highlights. Why on earth would you want to compress highlight detail, which is already an issue with digital photography? Link to post Share on other sites More sharing options...
Jamie Roberts Posted April 23, 2007 Share #166 Posted April 23, 2007 Another way to look at this is to look at levels per exposure zone (or stop). The way that it plays out for an M8 DNG versus 12- and 14-bit linear encoding is: Levels per zone Zone Leica DNG 12-bit 14-bit 0 7 4 15 1 4 4 16 2 4 8 32 3 7 16 64 4 9 32 128 5 14 64 256 6 18 128 512 7 27 256 1024 8 37 512 2048 9 54 1024 4096 10 74 2048 8192 Bottom line is that in zone 0 (deep black), you get 7 levels vs. 4 levels as the LFI article states, in zone 1 the same, and from there on, an M8 DNG has fewer levels that a 12-bit file. However, 7 levels vs. 4 in very dark blacks is an advantage that is marginal at best, given that that is also where any noise is. Sandy Sandy--are those the levels mentioned in LFI? Is there an online version of the LFI article somewhere? Something here is still fishy fishy fishy. You're saying in the midtones the m8 only has a maximum of 37 levels? And only 54 in the upper-mid-tones? And no more than 74 levels in the highlights? Something there is not right. Doesn't a common 8bpp JPEG contain 256 levels? Are you saying a Leica M8 DNG has less tonal depth / DNR than a JPEG? And still I can't get an M8 RAW file to band in blue skies, which would be the obvious result here if those levels were accurate... Does the non-linear nature of the encoding come into play here at all? (personally, if they were going to change the compression scheme, I'd like them to go to the more normal lossless compression of other RAW files. Yes, the files would vary in size). Link to post Share on other sites More sharing options...
ho_co Posted April 23, 2007 Share #167 Posted April 23, 2007 Advertisement (gone after registration) Sandy--are those the levels mentioned in LFI? Is there an online version of the LFI article somewhere? Jamie-Rubén's graph is slightly different from the one in LFI. The magazine shows the M8's performance as being a series of steps, horizontal lines connected to sloping ones connected to the next horizontal one. I don't believe the article is available online, but you might want to contact Michael Hussmann, the author of the article and a member of the forum. (Rubén's graph also differs from that in LFI in that the magazine shows only 14 stops' worth of data, while Rubén's continues to the right beyond that point. Also, Rubén's plot doesn't show the M8 line reaching the x axis at the 2 value, and the magazine's version begins at the x axis value of 1 instead of 2.) Doesn't a common 8bpp JPEG contain 256 levels? Are you saying a Leica M8 DNG has less tonal depth / DNR than a JPEG? Yes, JPG contains 256 levels. If you add the values in Sandy's second column, there are also 256 levels in the M8's 8-bit storage scheme. But that is how many boxes there are, and the values stored in the boxes extend to the full 14-bit range of the camera. And still I can't get an M8 RAW file to band in blue skies, which would be the obvious result here if those levels were accurate... Does the non-linear nature of the encoding come into play here at all? You got it: The full range of a RAW file (0-16383) is broken into 256 steps. To do this, each range of values is set equal to a single value but non-linearly: For example, all brightness levels from 6889 to 6972 are set equal to a single value; while all brightness levels from 1225 to 1261 are set equal to a single value. Part of the reason for lack of banding is that the full range of values to 2 ^ 14 is represented; another part is the interpolation of values necessary with a Bayer-pattern sensor. This latter point is probably the main reason for the excellent results of the camera despite its lossy compression. --HC Link to post Share on other sites More sharing options...
Jamie Roberts Posted April 23, 2007 Share #168 Posted April 23, 2007 {snipped}You got it: The full range of a RAW file (0-16383) is broken into 256 steps. To do this, each range of values is set equal to a single value but non-linearly: For example, all brightness levels from 6889 to 6972 are set equal to a single value; while all brightness levels from 1225 to 1261 are set equal to a single value. Part of the reason for lack of banding is that the full range of values to 2 ^ 14 is represented; another part is the interpolation of values necessary with a Bayer-pattern sensor. This latter point is probably the main reason for the excellent results of the camera despite its lossy compression. --HC Thanks Howard... So if I have this right, the M8 has as its compression scheme "256 buckets" of values from the original 14bpp file? And, as we noted before, some are approximated / averaged to single values? IOW, you can't really compare normal RAW 'levels' here because the M8s are non-linearly compressed from the original 14+2 bpp sensor (and the compression emphasizes shadow detail--which is exactly where it needs to be for low-light shooting or pushing from lower ISOs, due to noise interference) Is Ruben then right about the M8 having less highlight detail than an 8bpp file? Where is the cut off to upper midtones? I really don't care if the specular highlights get averaged; this would make them smoother (which is indeed my experience with the M8) Those of use coming from Canon RAW 12bpp files regularly draw some pretty funky curves from their raw files to get visible detail in the upper midtones--lots of levels near white don't really count for very much when you put the normal toe and shoulder onto images for print... I'm a bit confused as to how the Bayer interpolation would enter into this. Are you saying that because there's a higher bit depth sensor to begin with, the interpolation is more accurate (and therefore subject to less banding?) Thanks again! This has fascinated me from the start, but the results I get in practice with the M8 are so much better at low ISOs than other digicams I couldn't figure the reason why... (they are a couple of stops worth better through the midtones--which get pushed around a lot in my workflow) Link to post Share on other sites More sharing options...
ho_co Posted April 23, 2007 Share #169 Posted April 23, 2007 So if I have this right, the M8 has as its compression scheme "256 buckets" of values from the original 14bpp file? And, as we noted before, some are approximated / averaged to single values? Correct. The algorithm rounds all values in a particular range down to the initial one: A pixel delivers a value in the range 0 to 2^14. The M8 multiplies that value by 4; takes the square root; and truncates to the integer portion only. IOW, you can't really compare normal RAW 'levels' here because the M8s are non-linearly compressed from the original 14+2 bpp sensor (and the compression emphasizes shadow detail--which is exactly where it needs to be for low-light shooting or pushing from lower ISOs, due to noise interference). That is precisely Leica's point of view. Is Ruben then right about the M8 having less highlight detail than an 8bpp file? Where is the cut off to upper midtones? Yes, the M8's tonal separation at the high end is less than that of an 8bpp file. Because Hußmann's LFI chart simply shows a range from 1 to 14 EV, I don't know where to break out upper and lower mid tones. (Rubén's chart http://www.leica-camera-user.com/digital-forum/20079-official-response-leica-laundry-list-8.html#post218440 is similar to the one in LFI. Where the M8's red curve is below the others, its ability to resolve tonal differences is better than the others'. Where the curves of the others are below that of the M8, they distinguish tonal differences better than the M8.) I'm a bit confused as to how the Bayer interpolation would enter into this. This isn't immediately obvious and it took me a while to begin to get a handle on it. The question here with the M8 is, "Okay. You're only giving me 256 different values, even if they do range from 0 to 16383. So how to I get smooth gradations?" As I understand it from Michael's article, it is largely due to the interpolation necessary with any Bayer pattern sensor. That is, each sensor pixel delivers a value for only one color, R, G or B. But each pixel on the screen or print needs all three colors; and these additional colors are determined by interpolating from the surrounding sensor pixels of the given color. Fifty percent of all green values and seventy-five percent of all red and blue values are interpolated. So far, that's the case with all cameras using sensors with Bayer pattern sensors. As I understand it, since the M8's stored values for these pixels are stepped--that is, two adjacent values will have a separation of between 2 and 398--the interpolation will produce smooth transitions by re-introducing something *similar* to the original value. That is, the original values are lost; but since the values stored may be fairly widely separated, the interpolated values will also be fairly widely separated. In addition, because of the compression algorithm the stored values in the darker portions will be very close to the original values; the importance of the interpolatiion comes more into play in the higher values where the algorithm creates bigger value separations. In other words, the interpolation works to smooth values more strongly where the values are further apart--or, to put it another way, where the smoothing is more needed. Hußmann's article (in LFI 2/2007 February) is extremely interesting. It is written in perfect "Leica-speak," mentioning the disadvantages of the M8 compression scheme but stressing its advantages and giving just enough of a technical basis for us to go figure it out on our own. It would be very worthwhile if he could publish just the graph from the article, since it differs in a couple ways from Rubén's. I hope I've understood the matter accurately and conveyed it adequately. I think some of the innovations of the M8 may be accidental discoveries, but they certainly work together to build a superb picture-taking tool and Leica and Hußmann have done an excellent job of explaining how all the pieces come together. --HC Link to post Share on other sites More sharing options...
george + Posted April 23, 2007 Share #170 Posted April 23, 2007 Yes, it is a good article which explains well how Leica produces such superb images with the scheme that they implemented. Link to post Share on other sites More sharing options...
ho_co Posted April 25, 2007 Share #171 Posted April 25, 2007 Guy-- Another thing to ask Leica about: Find another source for the charger. I know this isn't as important as other issues, but it also doesn't take as much effort from within the company. Show them the modifications Mark Norton made on his M8 charger and just raise the question whether they couldn't find a better supplier. Just an idea. --HC Link to post Share on other sites More sharing options...
Guest guy_mancuso Posted April 25, 2007 Share #172 Posted April 25, 2007 I did put that on the extra options list. It is a brick Link to post Share on other sites More sharing options...
psquared Posted April 25, 2007 Share #173 Posted April 25, 2007 Does anybody else find it too easy to delete a picture inadvertently? http://www.leica-camera-user.com/digital-forum/22695-m8-too-easy-delete-files-error.html#post238289 Link to post Share on other sites More sharing options...
rosuna Posted April 25, 2007 Share #174 Posted April 25, 2007 The graph wasn't mine, but it was made only for illustrating the basic idea of the non-linear 8-bits compression. The small differences with the original graph are meaningless. I find the DMR files better in terms of color, sharpness and tonal variation than the M8 files. I don't know if tonal compression explains this or not, but 255 tonal differences per channel is even more than the 683 tonal compression applied by Nikon to the NEF files (9,5 bits or so). However, Nikon allows the option of non-compressed files. I would feel more comfortable with a compression to a bit wider gamut of tonal variations (non-linear 10bits, for instance). I know a photographer (Vergara, Barcelona) that uses the M8 and the DMR. He will make some comparative tests (he's busy now). I recommended him to take pictures of clouds, or white cars (other suggestions?). We are discussing this subject in another forum (Spanish language): Leica m8: primeras imagenes - Página 34 - OJODIGITAL Leica m8: primeras imagenes - Página 35 - OJODIGITAL This is the web of Vergara's company: Fotografia amb Leica DMR The inspection of the histograms of preliminary photographs (wine bottles in studio) shows how much the M8 compresses the tonal range. The information recorded are represented by the vertical bars. When you go back to a 16-bits space the gaps are filled, but the resulting shape isn't smooth. Compare it with the histogram of the DMR files. The DMR provides the best histogram. I don't know if this is relevant in practice, but the DMR files look much better to my eyes. The third camera is the Canon 5D. I am waiting for results from a comparative test DMR-M8, but the extreme compression of the tonal variation in the DNG files of the M8 is an important subject to investigate. Link to post Share on other sites More sharing options...
rosuna Posted April 25, 2007 Share #175 Posted April 25, 2007 The following graphs aren't mine. They were made by "GUI", a forum member of "Ojo Digital". They are two histograms of a segment from the shadows of two different photographs, from the Leica M8 and the Canon 350D. Leica's compression scheme is good enough for the shadows in theory (a bit better than linear 12bits in some parts, a bit worse in others), and the histograms confirm this. They illustrate how the compression works. The smooth shape is the resulting histogram in a working 16-bits linear space. The vertical bars are the information stored in the RAW files. In some sense, when you open the RAW file, you "interpolate" the information of the smooth histogram from the bars. Welcome, dear visitor! As registered member you'd see an image here… Simply register for free here – We are always happy to welcome new members! Link to post Share on other sites Simply register for free here – We are always happy to welcome new members! ' data-webShareUrl='https://www.l-camera-forum.com/topic/19830-official-response-from-leica-on-laundry-list/?do=findComment&comment=238061'>More sharing options...
rosuna Posted April 25, 2007 Share #176 Posted April 25, 2007 The problem could appear in the mid of the histogram and, specially, in the highlights. The Canon 350D stores in 12bits space without compression (linear), so it has 4096 tonalities. This is the graph for a zone in the middle (from 2048 to 2815) from the Canon 350D: Welcome, dear visitor! As registered member you'd see an image here… Simply register for free here – We are always happy to welcome new members! Link to post Share on other sites Simply register for free here – We are always happy to welcome new members! ' data-webShareUrl='https://www.l-camera-forum.com/topic/19830-official-response-from-leica-on-laundry-list/?do=findComment&comment=238067'>More sharing options...
rosuna Posted April 25, 2007 Share #177 Posted April 25, 2007 We know, from the graph of LFI or sandymc's table, that the M8 has information losses in the middle of the histogram. http://www.leica-camera-user.com/attachments/digital-forum/31887d1175473174-official-response-leica-laundry-list-8_bits.gif Levels per zone Zone Leica DNG 12-bit 14-bit 0 7 4 15 1 4 4 16 2 4 8 32 3 7 16 64 4 9 32 128 5 14 64 256 6 18 128 512 7 27 256 1024 8 37 512 2048 9 54 1024 4096 10 74 2048 8192 I find the comparison with 12-bits files unacceptable from a methodological point of view, because Leica's A/D converter has an internal output of 14-bits. This should be the comparison point of reference, not 12-bits. I don't mind how the M8 compares with 12-bits cameras. I want to know how it compares with uncompressed 14-bits files or uncompressed 16-bits files (DMR from Leica). Even if the 8-bits-compressed files from the M8 don't look too bad compared with linear 12-bits files, this is meaningless to me. The LFI graph would look much worse for the M8 files if the comparison point of reference is an un compressed 14-bits file. Anyway, this is that Leica does. Well... Welcome, dear visitor! As registered member you'd see an image here… Simply register for free here – We are always happy to welcome new members! Link to post Share on other sites Simply register for free here – We are always happy to welcome new members! ' data-webShareUrl='https://www.l-camera-forum.com/topic/19830-official-response-from-leica-on-laundry-list/?do=findComment&comment=238071'>More sharing options...
rosuna Posted April 25, 2007 Share #178 Posted April 25, 2007 The hightlights are more severely compressed in the M8 files. The first graph is from the Canon 350D (4096 to 4863 tones in the 16-bits space): Welcome, dear visitor! As registered member you'd see an image here… Simply register for free here – We are always happy to welcome new members! Link to post Share on other sites Simply register for free here – We are always happy to welcome new members! ' data-webShareUrl='https://www.l-camera-forum.com/topic/19830-official-response-from-leica-on-laundry-list/?do=findComment&comment=238073'>More sharing options...
rosuna Posted April 25, 2007 Share #179 Posted April 25, 2007 The second graph is from the M8 (same tones): Welcome, dear visitor! As registered member you'd see an image here… Simply register for free here – We are always happy to welcome new members! Link to post Share on other sites Simply register for free here – We are always happy to welcome new members! ' data-webShareUrl='https://www.l-camera-forum.com/topic/19830-official-response-from-leica-on-laundry-list/?do=findComment&comment=238074'>More sharing options...
rosuna Posted April 25, 2007 Share #180 Posted April 25, 2007 The original output of the M8 would look better than the Canon 350D's output (14 vs 12bits transposed to a 16bits space). The "input" histograms of the M8 looks worse than those of the Canon 350D because Leica applies a compression algorithm, and this procedure implies losses of tonal information. The "output" histograms, after distribution into a 16-bits space are similar (smooth shapes), at first glance. The "input" histogram of the DMR looks perfect, just because the information is stored in a 16-bits space (no vertical bands and interpolation). There are two separated questions here: 1) do you accept any lossing scheme? (you win smalle files and faster operation of the camera); 2) if you accept it, where is the limit? My response to the first question is yes, I would accept some losses if there are benefits. My response to the second question depends on the practical consequences. I don't want to loss visible image quality. The point is that I see a better quality in the DMR files. Is it due to the compression scheme or there are other reasons? This is a typical histogram from the DMR (from shadows to highlights): Welcome, dear visitor! As registered member you'd see an image here… Simply register for free here – We are always happy to welcome new members! Link to post Share on other sites Simply register for free here – We are always happy to welcome new members! ' data-webShareUrl='https://www.l-camera-forum.com/topic/19830-official-response-from-leica-on-laundry-list/?do=findComment&comment=238081'>More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.