|
|||||||||
|
Thread Tools | Search this Thread |
July 23rd, 2006, 03:07 AM | #31 | |
Major Player
Join Date: Aug 2004
Location: Kangasala, Finland
Posts: 445
|
Quote:
The worst example of my own XL H1 HDV footages is www.luontovideo.net/Hawk owl.tif and http://www.luontovideo.net/Great spo...oodpeckers.tif is about the same quality. As they stand, one could not edit these to a TV program to be shown by a national TV station. Still, I think, these footages can be saved by proper postprocessing. After these shots, I spent a considerable amount of time adjusting the custom presets, and the images indeed became (much) better. For example, www.luontovideo.net/Ravens.tif Now, although seldom mention on this forum, the XL2 does suffer for similar kind of problems. Putting the camera in extreme conditions will reveal the problems. For an --this time I guess real CA-- example, see www.luontovideo.net/Willow tit.tif If needed I can show much more examples of the XL2 as well. In any case, I think A.J. has a wisdom when he says one has to accept compromises. Although the marketing machinery lets us understand digital videocameras makes everything only better, this is not quite "the thruth and nothing but the thruth". Often, or should I say in most or in major cases, yes, digital video cameras are amazing compared to the old analog devices, but there is also a cost for the use of compressors etc. Steven Rosen mentioned some time ago on this forum, that the old analog videos seem to have only little artifacts although otherwise they did have shortcomings. That's the same impression I get when looking back to my old Betacam tape archive. The first aid of "red/green fringing, CA, border effect" --what ever one liked to call it-- were a proper postprocessing filter. Should not be imposible to code a filter to remove the bit annoying effect of XL H1 and it's standard 20x lens. Perhaps Cineform could make a one? |
|
July 23rd, 2006, 05:09 AM | #32 |
Trustee
Join Date: May 2003
Location: Atlanta GA
Posts: 1,427
|
I understand everyone's concern with C.A. but think Dave listed the problem and solution all at once. The XL-H1 is a great CAMERA for (indie) film and TV. The VF blows, but once you get past that (or get an HD-SDI on-board monitor or FU-1000) it's a great camera. Dave agree's other wise he wouldn't post as such at the end of every one of his posts.
BUT, the lens isn't good enough for big screen work. So it goes. Personally I tried to buy my camera without the lens (still isn't an option) because I love the 35 adapters, I love the DOF and I like being able to use more intechangable lenses. That being said how many people out there are using the 20x lens with the realistic notion that it will be blown up to 35? It's the same old story, no one is satisfied with anything, if they had given us a lens with better C.A. (or less C.A. or whatever) it would have cost more, and people would have said "I'm not spending 10k on an HDV camera". But if they had dropped the price to 7500 and not included a lens (so you could use the 1500 on a 35mm adapter and some lenses, or at least the rental of one...) people would have said "I'm not spending 7.5k for a camera without a LENS!" (even though it does have an HD-SDI output) For the love of god, HD is not Film, HDV is not HDCAM. All of these formats can be tweaked to look very good, and/or very bad. Dave has found a solution, I find it SLIGHTLY odd that in order to reduce the C.A. on a lens, you have to shoot through that lens and ADD more glass to it, but if it works for him perhaps that's something everyone else should try. For me the Mini 35 or the Letus seems to be a better option, but that's me. So the lens was a trade off between price and performace, has anyone priced true HD lenses? I'll go one step further and predict the future. When the 6x auto lens comes out (when I saw it at NAB-I know, it was a mock up- it looked AUTO not Manual, NOT like the 16x, or my favorite xl lens thus far the 14x) there will still be C.A. issues, but it'll be better then the stock lens, or at least wider and everyone will run to it and some will buy it then someone will point out something wrong with it and then there will be another thread just like this one. Certain people will weigh in and others will disagree, and at the end of the day no one's opinion will really change. I don't mean to belittle a problem or anyone's opinion, but if the lens is only going to be an issues on a 40ft screen, uh, don't use it. Use an adpater, problems solved, if it works for TV then use it for TV. P.S. the 20d stock lens stinks, put a respectable lens on there (24-70mm f/2.8L, 85mm f/1.2L II) an tell me that canon glass stinks... And for the record, my HDTV is 1920x1080, I have made the money back on my camera (though not all of the accessories) so my buyers remorse has almost past and I'm ready to date again, and most of my work is with the larger format HD cameras (Sony F900, Some Varicam, etc.) not film, so that's my frame of reference. An my backpack's got jets...
__________________
I have a dream that one day canon will release a 35mm ef to xl adapter and I'll have iris control and a 35mm dof of all my ef lenses, and it will be awesome... |
July 23rd, 2006, 08:41 AM | #33 |
Major Player
Join Date: Jun 2004
Location: McLean, VA United States
Posts: 749
|
This has certainly been a most intersting thread and I wonder if any conclusions can be drawn at this point. I spent the morning looking at lots of the images you guys have posted and I must say I saw just about every kind of defect that video is heir to. Things like plain old aliasing (moire), color noise, DCT roundoff (tiling), chroma/luma crossover, limited chroma bandwidth, "mosquito" effect and artifacts caused by sharpening (halos, color noise) and yes, chromatic aberration as well. I was most interested to compare the frames from still cameras with "no" chromatic aberration to those with "unacceptable" levels only to find that they had comparable levels (about 1 pixel at frame edge). From this I conclude that designers shoot for 1 pixel at the frame edge and in this sense I guess I'd have to say the 20X is OK. That they can hold it to 1 pixel in a lens this inexpensive with 20X zoom range is amazing to me. As I don't want to be thought of as an apologist for Canon let me say that I share most peoples' views about the clumsiness of the thing and difficulty in using it in manual mode. I'd much prefer a 'real' lens (as long as it has image stabilization).
So I guess my conclusion at this point is that there are things about this camera you don't like and that's fine because there are, as Chris put in the Texas shootout summary (I think it was), there are severe limitations in all the sub $10K HD cameras. I agree wholeheartedly with this statement. What I don't agree with is that a magic lens will solve the problem because some of the problems that you have complained about aren't caused by the lens. I suspect that what may be responsible for many of the problems you are seeing is sharpening. What sharpening does is enhance edges by boosting the video level on the bright side of a transition and lowering it on the dark side. As you move away from the edge the video level transitions back to the level recorded by the CCD. This process puts halos or bands around objects which are certainly noticeable if you are looking for it. There is an appreciable gain in perceived sharpness when you do this is that's why they do it. It's an engineering trade space. People used to scream bloody murder about the aliasing in the XL2. There are, on this board, lots of posts with a similar tone to that of some in this thread - that Canon cameras have unacceptable levels of Moire, that there is no excuse for this.... No one seems to complain about aliasing with the XLH1 which causes me to think that Canon put a blur filter in the optics to control it (and I think they did). As a result the picture is softer than it would be without the antialising filter which would cause people to complain about the resolution of the camera. The fix: sharpening. But nothing comes free. Sharpening introduces the halos and noise I mentioned earlier. Wouldn't it be ironic if Canon responded to complaints of aliasing by low pass filtering and sharpening only to be hoist upon their own petard in this same area of the design? If my thesis is correct there is a simple fix. Turn down sharpening and see if you like the result better. Halo's will go away but the picture will not be quite so sharp. Dave commented that he liked the result with the 35mm adpter. The ground glass in these adapters reduces the image sharpness somewhat - perhaps enough to offset the sharpening algorithm in the camera. Perhaps reduced sharpening will do the same. Maybe I am right about this, maybe not. What does it cost to try it? My opinion of the XL-H1: It's miraculous. At the price it's FM! It will put an image on my plasma TV which is better (lots better) than much of (but not all of) what is broadcast. In many cases it's images are better than those of movies presumably shot on 35mm film. It's images are sharper than what I see in the movie theatre in some cases. Perhaps I am more forgiving than some of you because as an engineer (on the creative side I'm pretty much limited to knowing you shouldn't pan and zoom in the same shot) I have more reasonable expectations on performance given the current state of the art. But you guys are right to keep demanding more. Eventually you will get it but the development cycle can be excruciatingly slow. |
July 23rd, 2006, 09:47 AM | #34 | |
Major Player
Join Date: Sep 2005
Location: Pleasanton, CA
Posts: 258
|
Quote:
Best, Christopher |
|
July 23rd, 2006, 11:29 AM | #35 |
Major Player
Join Date: Aug 2004
Location: Kangasala, Finland
Posts: 445
|
This is, indeed, an interesting and useful thread. Thanks for everybody.
A.J., I tested immediately what you suggested. The frames are here: www.luontovideo.net/Sharpness 0.tif www.luontovideo.net/Sharpness -3.tif www.luontovideo.net/Sharpness -6.tif www.luontovideo.net/Sharpness -9.tif But, also for my surprise, the visual impression is that decreasing sharpness will make the green border line more visible. Just compare the vertical lines between Sharpness 0.tif and Sharpness -9.tif. Second, while a took the footages, the light turned towards left. Only a moment after taking these images I could no longer reproduce the green line between wall and stairway on the right hand side of the image. This seems to confirm what I've observed earlier: light reflecting back to the lens make the situation worse. (... Well, if there was no light reflecting back to the lens, there were no problems, but the image would also be completely back. So, let's say shiny objects such as metal pipes, snow fields or white walls generate difficulties when the light source is behind the camera.) |
July 23rd, 2006, 11:49 AM | #36 | |
Obstreperous Rex
|
Quote:
|
|
July 23rd, 2006, 01:28 PM | #37 | |
New Boot
Join Date: Jul 2006
Location: Winnipeg, Canada
Posts: 17
|
Quote:
|
|
July 23rd, 2006, 01:49 PM | #38 |
Obstreperous Rex
|
The Canon 18-55mm EF-S is the stock lens bundled with the Rebel XT and 20D. I have one. It's pretty much a "freebie" lens in that it has a market value of less than $100. Now it's adequate for what it is, especially considering what it costs, but it is by no means any measure of Canon's true capability with optics. Any other EOS lens is much better than the 18-55mm EF-S.
|
July 23rd, 2006, 02:27 PM | #39 | |
Major Player
Join Date: Sep 2005
Location: Pleasanton, CA
Posts: 258
|
Quote:
Best, Christopher |
|
July 23rd, 2006, 02:53 PM | #40 |
Major Player
Join Date: Jun 2004
Location: McLean, VA United States
Posts: 749
|
Lauri,
I looked at (analyzed) your images and here's what is going on. Coming out of the CCD red and blue are sampled at a lower rate so that while the transition between the dark area and light area (between stairs area and wall at the right of the picture) in the green channel takes about two pixels it takes about 4 pixels for the red and about 6 for the blue. Thus the first few pixels after the transition (moving to the right) look green in the raw image (no sharpening). When you apply sharpening the green transitions in about a pixel's width as does the red while the blue takes about a pixel and a half. Thus the green band after the transition is narrowed. Note that the blue transition is shifted about a pixel to the right of the red and green transitions. This IS chromatic abberation. The reason the system is designed this way is because you aren't supposed to to be able to see color shifts of these magnitudes at these rates unless you sit close enough to be able to clearly distinguish individual pixels which in theory you aren't supposed to do. The fact that you can see them either says that you are looking too closely or that you have unusually sensitive color vision either or both of which is possible. Dave wondered why most people here can't see these things. I suspect the answer is that while the whole of color science is based upon the "average observer" you two (and doubtless others as well) are not average. I'm at the other end of the spectrum. To see the fringe clearly at one display pixel per image pixel at comfortable viewing distance I have to boost saturation appreciably (yes I'm a protanope). So in this case the "problem" is simply that of reduced chroma color bandwidth. It is a shortcoming of the system which is designed in. It should be better with 4,2,2 sampled data from the SDI port (I assume this was an HDV capture) but still not as good as it would be with 4,4,4. With respect to CA we note that while it is there its effect is masked by the color bandwidth limitations. A better lens would not fix it. As to why this goes away when the angle of the light shifts all I can say is that we now have a whole new situation. Perhaps the contrast is less across that transition when the light is more angled and maybe that is enough to explain it. Thanks for posting those images. Very interesting! And thanks for your kind words as well. They are appreciated. |
July 23rd, 2006, 03:08 PM | #41 | |
Obstreperous Rex
|
Quote:
|
|
July 23rd, 2006, 08:32 PM | #42 |
Regular Crew
Join Date: Mar 2004
Location: sherbrooke (Quebec) Canada
Posts: 108
|
Chris,
from what i've seen so far, from the res charts, the manual lense looks a lot more "solid". There is almost no CA, and the res is as good as the stock lense. I think the 20X lense does create some CA. The 3X is also a lot cleaner than the 20X to my eyes. |
July 24th, 2006, 12:41 AM | #43 | ||
Major Player
Join Date: Aug 2004
Location: Kangasala, Finland
Posts: 445
|
A.J.,
Thanks for your excellent analysis. All this makes now a lot of sense. Quote:
Quote:
Now, that the issue is understood makes it clear how the custom presets can be exploited to remove the effect -and this vindicates A.J's hypothesis; Instead of setting sharpness I made a new experiment with Noise reduction 2, which I set to "high". The frames are here www.luontovideo.net/Neutral.tif www.luontovideo.net/NoiseReduction2.tif and the green border between the wall and stair way is less noticeable. (I also use gamma set to Cine1, which I've found to help with the border effect. The reason is clear: contrast becomes lower.) All this makes me now to suspect that the EF-adapter is a kind of low pass filter which acts bit like Noise reduction 2 softening the image. This explained i) why the images produced with EF-lenses seem to have less contrast as if they were slightly grayish compared to the standard 20x lens, ii) why people report "there is less CA with EF-lenses" and also iii) explained why Canon say the EF-lenses are not HD-compatible. The visual impression of XL H1 + EF lenses is though still very pleasing. The moral is, judging images based only on visual examination is very difficult and easily leads to misconclusions. Still, would like to emphasize, the final conclusion is to be made. Last edited by Lauri Kettunen; July 24th, 2006 at 03:04 AM. |
||
July 24th, 2006, 05:55 AM | #44 |
Major Player
Join Date: Jun 2004
Location: McLean, VA United States
Posts: 749
|
Chris,
I'd say we should refer to Lauri's wall problem as the "chroma undersampling effect". Note that I'm using "effect" rather than "problem" because it is there intentionally as part of the design and there are a couple of ways (that we know of at this point) to get around it: increase sampling to 4:2:2 (use SDI), turn on noise reduction, lower contrast (either through lighting or gamma) try to keep strong verticals out of the composition, try to avoid transitions with the bright side on the right... Or we could call it the "Lauri's wall effect". Whatever we decide to call it I don't think that is the end of the story because I don't think this phenomenon covers all the artifacts that people have been talking about in this thread. The fact that this particular situation was not caused by chromatic aberration does not mean that the XL-H1 doesn't exhibit some chromatic aberration with the stock lens. It definitely does. Also aliasing, DCT tiling, haloing, and all the other things I mentioned in an earlier post are still being called chromatic aberration by some. A.J. |
July 24th, 2006, 08:37 AM | #45 | |
Major Player
Join Date: Aug 2004
Location: Kangasala, Finland
Posts: 445
|
Quote:
My background is in scientific research, and think it's never a good idea to name something after somebodies name. |
|
| ||||||
|
Thread Tools | Search this Thread |
|