Personal View site logo
GH2 AVCHD Encoder Motion Testing
  • There is a lot of static testing going on. I wonder, though, whether the real advantage of higher bitrates manifests itself in dynamic scenes more than static ones. Therefore, it seems logical to set up some sort of standardized test for resolution vs. movement. Something like shooting a resolution chart while panning slowly (at a fixed rate) might show whether there is improvement in dynamic scenes.

    Shutter speed will be an important consideration, as the slower it is the more blurring (and the less detail) there will be. At fast shutter speeds minute detail will change considerably between frames.

    With a static scene I frames dominate, and B and P frames don't play much of a role so quality improvements concerning those won't show up.

    I'm going to play with this a bit. In the meantime - any thoughts?

  • 43 Replies sorted by
  • Shame no-one still has record players. You could make a nice round 12" test chart and rotate it at a known speed (33-1/3) and there you are. Difficult to otherwise think about how you could do a standardized speed.
  • I have a motorized pan-tilt head, so I can do my own testing with it. The record player is a good idea, except I think the rotation will be too fast to produce usable results. The absolute pan speed is less important than doing it approximately the same speed for each test. We may find, for example, that resolution drops considerably between I frames with low bitrates, but doesn't with higher bitrates. It's the resolution in I frames vs. B and P frames that seems important to me - especially with high shutter speeds.

  • Hi Chris,

    I was being slightly facetious about the record player, but actually a spiral printed on a circular bit of paper on the turntable would produce a quite slow movement (depending on the pitch of the spiral). But I absolutely agree the importance of having a moving target of some sort. I wonder how many of us have record players anyway...

    ...or some sort of rotating pattern on screen? Either way they would be pretty consistent speed-wise.
  • I've tried the screen thing - it doesn't really work. First, resolution is low, and there are moire problems. Actually, I'm not sure pan speed is all that important, as long as it is slow. I think what we're looking for is I frame quality vs. P and B frame quality in motion. Results should be pretty visible no matter what the pan speed is - so long as it's reasonable.
  • How about standardizing on a classic analog metronome:

    For high-bitrate motion testing, I typically use a 1/125-second shutter speed in sunlight. However, it's difficult to maintain consistent lighting between takes. With indoor fluorescents, consistency is not a problem, but flicker-free shutter speed is restricted to 1/60-second. Since PAL would require a different shutter speed, I'm not sure what the best lighting solution would be?
  • The metronome sounds good. I assume you would tape a small res chart or some other pattern picture to it. I think we want there to be a lot changing in the image field.

    I'm using incandescent lighting.

    I suspect using several shutter speeds would be good - all the way up to 1/500. The faster the shutter, the more extreme the fine detail changes.
  • On my record player the grooves all move at different speeds. Well, technically, it is one groove, of course.
  • Come to think of it you could just hang a res chart from the ceiling on a string and swing it - cheaper than buying a metronome. If it was swung from the same distance at the start of each test the results would be pretty consistent. Actually, two strings would be better; that way you wouldn't have to worry about twisting, etc...
  • DrDave,

    Are you suggesting that other people have record players where all grooves move at the same speed? Unless, of course, you're alluding to those of us who own Edison drum record players.
  • The convenient thing about a metronome is its repetition rate is calibrated and easily adjusted. For me, the most challenging aspect of fine-tuning patch settings on the GH1 has been the problem of producing consistently repeatable test shots. Thanks for starting this topic, it's prompted me to consider setting up a more systematic test rig for the GH2.
  • I just bought a metronome. I want to see how it affects codec break-up with high shutter speed. I have a powerful video light to use with it as well.
  • Here's a simple res chart I built. Align the arrows to match up with the vertical frame size. I added diagonal items because I've noticed that strange things happen with diagonal lines sometimes. Print it, glue it to a card, and hang the card with two strings from the ceiling. Then swing it so it goes from frame edge to frame edge. With strings of 1m the swing period should be about 2 seconds. The image is 8"x10" at 720dpi.

    [EDIT] Use the newer file below...
  • What I mean by "from frame edge to frame edge" is to swing it so the target image just swings out of view.

  • Chris--a standard LP has two grooves, one on the back and one on the front. Any object placed on the outer part will move faster, proportionally, than one on the inner part. The groove itself can be said to move at one speed if considered to be one single groove, or at different speeds if it is imagined that it is many grooves. A cylinder is of course a superior design, as the groove moves uniformly, however this design was more expensive to mass produce and has variance across a frame for video purposes. A metronome of course has the same issue. A large hedge trimmer would have uniform motion, as opposed to a drum which would be no problem at a single point, as in an Edison device, but which would present similar issue in a cross-section when videoed. A hedge trimmer is better for uniform video than a weed whacker, basically, although both are interesting. Perhaps for this purpose it does not matter, or perhaps the velocity differential, which can be easily calculated, is good for finding a specific speed at which the codec breaks down.
  • Here's an improved res chart. It includes a star target so we can see symmetry.

    288 x 360 - 32K
  • DRDave,

    I think the variable rate swing (oscillation) is actually a good thing as it will allow for a spectrum of change rates. One thing comes to mind, though. Instead of swinging it to outside the frame, maybe it would be better to swing it from edge to edge inside of the frame. Experimentation will tell...

  • Vitaliy,

    Go ahead and remove the old chart file if you want.
  • we could mount a cardboard with chart on it to a dc-motor, set with an adjustable power supply slower or faster
  • That's a good idea too. I was just trying to come up with a solution that doesn't cost anything so more people could participate.
  • The idea of a standardized testing setup is great. Just some ideas, Same lighting steps, same distance, as close as the same lens as possible, same fstops, etc.... Then we can post/graph/grid the results based on some type of quality indicator.
  • I suppose you could generate an image that would correct angular velocity, to check the evenness through the frame
  • I'm not sure aperture, lens length, etc... makes any difference. Pretty much any lens configuration will max out 1080 resolution. It is important, however, to have the target the right size in the viewfinder. I plan to do most of my testing with the 20mm lens. What we're looking for is relative resolution between frame types; absolute metrics concerning lenses etc... is another issue. As for how fast the image is moving - well, we are shooting video after all, so speed should be easy to measure. Also, you could calculate image speed vs. angle from center as long as you know what the extremes are.

    Why would an image that corrects angular velocity be necessary? All we want to see is how well detail holds up in P and B frames vs. I frames, and whether higher bitrates improve the temporal behavior of the codec. I wonder, for example, whether even if a higher bitrate doesn't improve static images, if it does improve dynamic ones. For example, if you look at an I frame and the immediately preceding frames have much lower detail then we know detail is being lost, at a higher bitrate they might look better even if the I frame doesn't. After several seconds it should be possible to have captured samples of images in all positions and with every frame type.

    I'll play with this a bit more and see if I can come up with a simple protocol.
  • guys, honestly! just remember your physics! :-)

    A pendulum's period (timing) is determined (basically) by its LENGTH. Settle on a standardized length, weight, release distance, and a chart. set up, shoot, repeat.
  • String length = 1 meter. Use video from when the target goes from edge to edge (inside) - you just have to be somewhat wider when you start. Weight doesn't matter (except the swing will last longer). Pick frames close to the center. Make sure the target isn't swinging in and out of focus (takes a little practice). I have some test results. I wasn't all that precise and the results are very revealing - stay tuned.

  • @cbrandin

    Try to test this with manually adjusted bottom bitrate for 24H mode (in testers section).
    Like at 90% of top bitrate.