First Impression; serious drawbacks

I have commented many times on this forum about this exact effect you are describing and have been howled down by many. The judder I have seen with the AF100 and GH1 for that matter, is nothing like the progressive "judder" I get on my Varicam and SDX900 in 25P. It is like an "intraframe" strobe. I believe it is directly related to the CMOS sensor/shutter on the AF100, GH1/2 and nearly all DSLR-related CMOS cameras. I agree 100% with your findings and would bet the farm that your CCD global shutter HPX3100 won't exhibit the same progressive-frame "juddering" you are seeing with the AF100. But as the song goes, "...two men say they're Jesus, one of them must be wrong."
 
Last edited:
nugat said:
What we need to see is is really slow movement of an object across the frame for more than 7 seconds. My Vimeo clip* has the window vertical border travelling 15 seconds (not 4.5 sec) across the frame, and it still judders. It is not due to coding issues either, because I can assure that it looks exactly the same on the plasma played back from PS3, AF100, in FCP, iMovie, re-wrapped in ClipWrap from .mts into .mov etc etc.

* http://vimeo.com/20429696
Click to expand...
I downloaded the MP4 file from Vimeo and imported it into a Premiere Pro CS5 25p sequence. At 100% magnification, I see a prominent flicker on the edges of the potted plant as it scrolls across the frame, both at 25p and single-stepping through the frames. I've seen this distracting effect numerous times on high-contrast scrolling shots I've taken in daylight, particularly when shooting at high shutter speeds.

As an experiment, I displayed the video at 200% magnification in Premiere Pro (with Full Playback Resolution). This dramatically reduced the flickering effect, though I could still detect small traces of flickering on the edges of the plant stalks. It's obvious to my eyes that Premiere uses a smoothing filter at 200% magnification that reduces the contrast on sharp edges.

I believe this flicker is a side-effect of the sharpening filter used in the demosaicing algorithm the camera uses to convert the RAW Bayer Filter output into RGB pixels. When an object with a high-contrast edge moves across the frame, its alignment with the image sensor RGBG grid varies slightly with each frame. As a result, the calculated RGB values of the pixels on the edges can vary noticeably from frame to frame. In the early days of low-resolution 3D computer animation, we used to call these edge effects "farkle" to distinguish them from the flicker due to low frame rates.

If you single-step through the video, you'll see that this edge flickering is only apparent in the instant that one frame is replaced by the next - it does not persist for the entire duration of each frame. In addition to the sharpening filter in the camera, the response time of the video monitor will have an effect on the flicker. Unfortunately, hi-res displays with faster response times (such as LCD's designed for gaming) will make the flicker appear more prominent. Slower displays will tend to smooth the high-contrast edges and minimize the flickering.

There are two approaches that may help suppress this flickering effect. When shooting high-contrast moving subjects, using slow shutter speeds and minimum in-camera sharpening should help smooth out the sharp edges. In addition, post-production smoothing filters can be used to selectively blend problematic edges into the background. I'm sure the gross improvement I saw using Premiere's 200% magnification could be done with far more skill and subtlety.
 
I should have added that although I see this judder, I just work around the issue and shoot in a style that doesn't make it objectionable. I don't have much faith in reducing the effect in post.
 
I just watched the first broadcast of episode 1 of a TV series we shot last year. It was shot with EX cameras, XDCAM HD and Canon 5D, all in 25P.
It contains "run and gun" docu sequences, extreme sport (wingsuitflying, skiing et),nature shots and interviews. We used dolly shots, steadicam, handheld, helicopter, you name it.
It looked fine, the way you would expect from 25P

I´vw watched 25P footage from another TV series (not broadcast, but over the HDMI played back from my AF101) I´m shooting right now on the same TV and the footage from the AF101 looks so weird/bad compared to the other cameras.
I should add that the TV is a 42" Panasonic plasma.

I´m only talking about the "judder" here. I have some very nice sit down interviews that looks great, but as soon as there is a lot of motion going on it looks like something is "wrong".
You still see some "judder" from the other cameras, but the "judder" looks a lot "harder" on the AF101
I have posted about this earlier in another thread.
I don´t have the camera with me to do anymore tests as it´s out on location used either by me or another cameraman
 
Off, so 1/50. It looks in away like high shutter or 90 degre shutter. (Private Ryan style)
 
LPowell, this is the first plausible hypothesis to what might be really happening. I hope it can be verified through further experiments. Thanks.

Meanwile here are new developments.

Firstly thanks to the generous offer of Fohdeesha, my "AF100 judder" clips* are now available on his FTP server.
Go to :
http://fohdeesha.com/data/other/public/
You'll see four clips/links for download. The top two "AF101PANLONG" is the longer 65 second/170MB series of panaromas of varying speed. The MTS container is the original camera file. The MOV container is the same file re-wrapped into Quicktime by ClipWrap (divergentmedia.com) . This does not change the sample structure of the original video file (no transcoding), it's the same codec with a different container that allows it to watch in QT.
You will need a strong machine to do it without problems, best a recent (2010-2011) multicore Intel Mac with Snow Leopard/ QT X and/or hardware acceleration. H264 is that demanding. Go to Divergentmedia site for an excellent tutorial .**
The bottom two clips at the FTP site is the shorter (25sec/70MB) single slow panaroma, again in MTS and MOV versions.

First I watched the clips on my home system (iMac i7 /8GB + EIZO cg243W coloredge in REC709 preset) and played from AF101 (original card/file) via HDMI onto my HD plasma. When watching on a computer display it's important to have the clip mapped 1:1 onto 1920x1080 as I noticed rescaling changes the strength of the phenomenon.
Today I projected the clip again from AF101 via HDMi and Panasonic PT3000 HD projector in the Cinema1 preset. Screen diagonal 120 inches (3 meters)

I found out the seeming "judder" (maybe it's the "farkle"?) depends on display type and viewing distance. It's strongest on the computer display (although the cg243 is also used as reference monitor for rec709 video), medium on the plasma, and weakest via projection. That also corresponds to the typical viewing distance: less than screen diagonal with the computer, more than a diagonal with plasma and projection. But the distance seems to be secondary factor to the replay/display method. What really made me wonder is the virtual absence of "judder/farkle" from my clips via projection. But in its place I noticed on the screen a strong chromatic aberration (greenish on one side of the window framing, purplish on the other). This is very unusual, as this lens (Zuiko SHG 14-35/2) is known for very low CA, virtually non-existent at f7. Any other explanation? Those greenish/purplish tints also seem to be "ringing" (changing intensity) with the movement. Is it in line with the explanation LPowell gave? (below quoted). Strangely enough the "CA"/"non-CA" is less visible on the plasma and I missed it first on the computer. There the phenomenon looks more like a typical judder, not a "ringing of CA artefacts".

That's where I am today with all this. LPowell's hypothesis is the most promising explanation , and my projection observation of disappearance of judder and appearance in its place of "CA ringing" the most intriguing--for me. I still plan to conduct comparative tests with AF101 and HPX3100 some time this month. Hopefully we are getting closer to an explanation.

* I placed the clips first at Vimeo with all setup parameters
http://vimeo.com/20429696

** http://www.divergentmedia.com/blog/fullpost/h264_decoding_on_the_mac

http://www.divergentmedia.com/blog/fullpost/choosing_postproduction_codec








lpowell said:
I downloaded the MP4 file from Vimeo and imported it into a Premiere Pro CS5 25p sequence. At 100% magnification, I see a prominent flicker on the edges of the potted plant as it scrolls across the frame, both at 25p and single-stepping through the frames. I've seen this distracting effect numerous times on high-contrast scrolling shots I've taken in daylight, particularly when shooting at high shutter speeds.

As an experiment, I displayed the video at 200% magnification in Premiere Pro (with Full Playback Resolution). This dramatically reduced the flickering effect, though I could still detect small traces of flickering on the edges of the plant stalks. It's obvious to my eyes that Premiere uses a smoothing filter at 200% magnification that reduces the contrast on sharp edges.

I believe this flicker is a side-effect of the sharpening filter used in the demosaicing algorithm the camera uses to convert the RAW Bayer Filter output into RGB pixels. When an object with a high-contrast edge moves across the frame, its alignment with the image sensor RGBG grid varies slightly with each frame. As a result, the calculated RGB values of the pixels on the edges can vary noticeably from frame to frame. In the early days of low-resolution 3D computer animation, we used to call these edge effects "farkle" to distinguish them from the flicker due to low frame rates.

If you single-step through the video, you'll see that this edge flickering is only apparent in the instant that one frame is replaced by the next - it does not persist for the entire duration of each frame. In addition to the sharpening filter in the camera, the response time of the video monitor will have an effect on the flicker. Unfortunately, hi-res displays with faster response times (such as LCD's designed for gaming) will make the flicker appear more prominent. Slower displays will tend to smooth the high-contrast edges and minimize the flickering.

There are two approaches that may help suppress this flickering effect. When shooting high-contrast moving subjects, using slow shutter speeds and minimum in-camera sharpening should help smooth out the sharp edges. In addition, post-production smoothing filters can be used to selectively blend problematic edges into the background. I'm sure the gross improvement I saw using Premiere's 200% magnification could be done with far more skill and subtlety.
Click to expand...
 
Last edited:
Perhaps you could ask the mod to merge your "25p trouble or is it just me?" thread with this one?
Any chance to post clips from your AF101 movie?
Cheers,

Yoak said:
I just watched the first broadcast of episode 1 of a TV series we shot last year. It was shot with EX cameras, XDCAM HD and Canon 5D, all in 25P.
It contains "run and gun" docu sequences, extreme sport (wingsuitflying, skiing et),nature shots and interviews. We used dolly shots, steadicam, handheld, helicopter, you name it.
It looked fine, the way you would expect from 25P

I´vw watched 25P footage from another TV series (not broadcast, but over the HDMI played back from my AF101) I´m shooting right now on the same TV and the footage from the AF101 looks so weird/bad compared to the other cameras.
I should add that the TV is a 42" Panasonic plasma.

I´m only talking about the "judder" here. I have some very nice sit down interviews that looks great, but as soon as there is a lot of motion going on it looks like something is "wrong".
You still see some "judder" from the other cameras, but the "judder" looks a lot "harder" on the AF101
I have posted about this earlier in another thread.
I don´t have the camera with me to do anymore tests as it´s out on location used either by me or another cameraman
Click to expand...
 
I'm fine if they merge the threads. How do I go about to do that?
All the clips I have "belongs" to the production so I can't post them.
The camera is out on location when I'm not using it. I will post a clip a soon as I have an opportunity
 
Just write them.

Yoak said:
I'm fine if they merge the threads. How do I go about to do that?
All the clips I have "belongs" to the production so I can't post them.
The camera is out on location when I'm not using it. I will post a clip a soon as I have an opportunity
Click to expand...
 
Nugat, here's another potential angle to consider. I noticed that you shot your Vimeo clip with an AF101 using an Olympus 14-35mm f2.0 lens. On the FourThirds.org site, one article describes how Zuiko Four Thirds lenses are specifically engineered to prevent oblique light rays from exposing the image sensor:

http://www.four-thirds.org/en/fourthirds/index.html

If this analysis is more than just a marketing pitch, it implies that Olympus and Panasonic lenses may have been designed to focus peripheral light rays in a different manner than traditional film lenses. Oblique light rays tend to reduce contrast in an image. Bending these rays to be more nearly perpendicular to the image sensor may contribute to the sharply-detailed "digital" look that's often attributed to these lenses. If so, it may also be enhancing the edge-flickering seen in your video.
 
Olympus creating the 4/3 standard wanted to make everything possible of the smaller (17.3x13mm) than full frame (36x24mm) sensor. That meant the glass was designed and built without compromise. It was almost as big as FF glass, but virtually free of distortions, aberrations etc. The top zoom line, SHG, was also exceptionally fast at f2 max aperture.
One of the features of the Zuiko Digital glass is near-telecentricity. It means the light angle of incidence is nowhere bigger than 7 degrees. The old film-style lens design was one of the main problems on digital sensors, the rays were progressively more oblique with the distance from the center. That was not a problem with film, which accepted a wide range of incidence angles. But with digital sensors, where photosite wells "do not see clearly" obliqe rays, it resulted with all kinds of aberrations, defects, loss of sharpness. One of the countermeasures was to place custom microlenses to refocus those oblique rays (eg. Leica M8 and M9). But much better solution was to create a new telecentric line of lenses--Zuiko Digital. It's highly unlikely that telecentricity causes artefacts on digital film sensors. If they were present, they should also strengthen with the distance from the optical axis, eg. loss of sharpness would be a telltale. Nobody signalled anything like that so far. Of course this can be easily verified with a set of different lenses, like panasonic lumix line built for m43. They are hardly telecentric, but another method of corrections was employed to improve performance-- in-camera and software manipulations. BTW, it is interesting if AF100 uses any kind of software corrections on its pictures. Does Gh1/Gh2 do it while in film mode? Anybody knows?
One way or another it'll be good to test and compare also this possibility. Truth is that Zuiko are world top glass as far as MTF representations. I haven't seen Zeiss Master Primes MTF curves, but from tests conducted it seems they reach 40-50lp/mm at 60-70% contrast across the frame. That's a similar figure to Zuiko Digital SHG. To put plainly MP's should also create problems on AF100 due to it's high contrast. Too good a lens for a body? Hardly likely, but who knows...


lpowell said:
Nugat, here's another potential angle to consider. I noticed that you shot your Vimeo clip with an AF101 using an Olympus 14-35mm f2.0 lens. On the FourThirds.org site, one article describes how Zuiko Four Thirds lenses are specifically engineered to prevent oblique light rays from exposing the image sensor:

http://www.four-thirds.org/en/fourthirds/index.html

If this analysis is more than just a marketing pitch, it implies that Olympus and Panasonic lenses may have been designed to focus peripheral light rays in a different manner than traditional film lenses. Oblique light rays tend to reduce contrast in an image. Bending these rays to be more nearly perpendicular to the image sensor may contribute to the sharply-detailed "digital" look that's often attributed to these lenses. If so, it may also be enhancing the edge-flickering seen in your video.
Click to expand...
 
Last edited:
Interesting, I shot mostly with the 14-35mm as well, but the other camera had a 14-140mm Lumix most of the time. We saw the problems on 2 different cameras
 
nugat said:
It's highly unlikely that telecentricity causes artefacts on digital film sensors. If they were present, they should also strengthen with the distance from the optical axis, eg. loss of sharpness would be a telltale. Nobody signalled anything like that so far. Of course this can be easily verified with a set of different lenses, like panasonic lumix line built for m43.
Click to expand...
I'm not suggesting that the Zuiko lenses' telecentricity may be creating digital artifacts - quite the opposite, actually. If the physics works as intended, these lenses are likely capable of producing more finely detailed, higher contrast images on digital sensors than legacy lenses from the film era. And that may be part of the problem with edge flickering in video sequences.

One of the more difficult challenges encountered in 3D animation is smoothly rendering a vector-based object moving across a 2D raster of discrete pixels. If each frame is rendered at the highest quality resolution, it can produce distracting image details that undermine the visual continuity of the object in motion. For optimal motion picture quality, it's often necessary to use motion-blurring and edge-smoothing image processing to actually reduce the amount of detail in the moving objects in a sequence of frames.

While the Four Thirds system was engineered to optimize digital image quality, it was designed and tested not for motion pictures, but for still-frame photography. None of the temporal artifacts that we see in videos that suffer from edge-flickering are present in the individual frames that comprise the video sequence - to the contrary, the details of each frame are rendered at the highest resolution possible. But when these individually digitized still images are displayed in sequence, the minute discrepancies in their finest details may create distracting discontinuities, particularly on the high-contrast edges of moving objects.
 
What detail setting do people shoot at? I never go above -5 and I too wonder if the detail enhancing algorithms are just too much. That said, I have still yet to see where someone has taken two different cameras, mounted them vertically above each other, and put lens on them to have the same FoV to a high contrast object to show the difference. I am not saying people are right or wrong, but words don't help in this situation. People need a frame of reference.

I shoot 24p at -5 detail on my AF100 and haven't see anything that looks wrong to me on my own footage. I've looked at some of the uploaded footage and nothing feels out of place. That said, what basis do I really have? There is so much perceptual difference going on it's really hard to say. That is what needs to be eliminated.

If you take multiple camera shots done EXACTLY the same, you can at least analyze the blurred edge of the motion. You can check the shift between frames on both cameras. You can get in close and try and determine what it is that you see. Then, you can show what is going on. Then people can help. As many know, I very much like to help out on a technical level as much as I can, but there isn't enough good/bad examples to go on. The word "judder" or "strobbing" could indicate different things in different parts of the image for some people. People are using these as if they are the same and I see them as much different, as they are.

I said this in the other thread, let's see some examples between two cameras! :)
 
Holdan got together with Alan Roberts, the guy who evaluates all the cameras for the BBC and one of the most adept technical guys on the planet, to explore this issue and get to the bottom of it. They concluded that there is nothing amiss and it's behaving exactly as it should. Holdan is looking to get the video online.

As for pans tested between two cameras -- I could do a test between a GH2 and AF100, or between an HMC40 and AF100, or between an HMC150 and AF100, or between 16mm film and AF100 (although obviously that would take a week or so to get the results back). Or between a Fuji Finepix stills camera in video mode, and an AF100. Or between an iPhone 4 and an AF100, or a Blackberry Storm and an AF100. Or between a webcam and the AF100. Those are about all the imagemaking devices I can think of that I have on hand. Which would you want to see?
 
You must log in or register to reply here.
Back
Top