How can I compare motion cadence?

[kye]

I have a few cameras that reportedly vary with how well they handle motion cadence.  I say reportedly because it's not something I have learned to see, so I don't know what I'm looking for or what to pay attention to.

I'm planning to do some side-by-side tests to compare motion cadence - can you please tell me:

1) what to film that will highlight the good (and bad) motion cadence of the various cameras, and

2) what to look for in the results that will allow me to 'spot' the differences between good and bad.

Thanks.

I'm happy to share the results when I do the test.  I'll also be testing out if there's some way to improve the motion cadence of the bad cameras, and what things like the Motion Blur effect in Resolve does to motion cadence.

[Towd]

I'm really interested in seeing the results of your test, and I've thought about this myself.  I think shooting something high speed that is repeatable would be interesting to see if there is any unusual timing-- like a recording frames at 60 hz, but then dropping frames and just writing 24.  Maybe shooting an oscilloscope running a 240 or 120 hz sign wave or something so you could measure if the image is consistent from frame to frame.

Another option for checking if frames are recorded at consistent intervals would be to use a 120hz or higher monitor and record the burned in time code from a 120 fps video.  This might be enough to see if frames are being recorded a a consistent 24 fps and not doing something strange with timing.  You'll probably have to play around with shutter angles to record just one frame.... or you could add some animated object (like a spinning clock hand) so you could measure that you are recording 2.5 frames of the 120 fps animation as would be correct for 180 degree shutter recorded at 24fps.  A 240hz monitor and 240fps animated test video would be even better as you could check for 5 simultaneous frames.  Anyway, just throwing out some ideas.

The only other things I can think of that could be characterized as bad motion cadence, might be bad motion blur artifacts, bad compression artifacts, and rolling shutter.   These might show up with a "picket fence" type test by panning or dollying past some kind of high contrast white and black strips.  At the very least it would be interesting to see if the motion blur looked consistent between cameras with identical setups.

[BTM_Pix]

This is what I've gleaned on the issue over the years...

To test motion cadence effectively you need to mount the two cameras on a dual mount tripod bar and ensure you have ,if not the same lens, at least the same focal length one on each.

For a test subject, you need something that can move of its own accord so you will see motion within a static framing but also that has enough movement to exit the frame so you can pan to track it and see the effects on a moving framing.

The ideal subject for this when it comes to testing motion cadence is, by common consent, a unicorn.

The ideal speed for the unicorn to be moving is from a slow walk up to a gentle trot.

Most unicorns are lazy but skittish, so a few party poppers will enable you to encourage this range of movement.

On the camera itself, make sure that the power draw isn't effecting motion cadence by using a fully charged battery and that you have mojo set to 10 or boost if your camera has it.

Depending on the brand of camera that you are using, you should also ensure that you have fully topped up your Koolaid levels.

😉

On a more serious note, this is a decent thread with some informed discussion about it and also contains a link to a side by side video between real film, Alexa, Red and the Sony F65.

https://cml.news/g/cml-raw-log-hdr/topic/16427778?p=Created,,,20,1,0,0::recentpostdate%2Fsticky,,,20,2,0,16427778

 

[kye]

Thanks @BTM_Pix - I'll keep an eye out for unicorns on ebay.

I think I've hit my first issue.  I've watched the video they linked and I see judder in all of them.  First attempt was my MBP running my 4K display with internal GPU, unplugging the external display my second attempt was MBP running only the laptop screen, third was plugging in my eGPU with RX 470.  All attempts displayed significant judder, both when playing the video fullscreen as well as at a 100% window (720p).

I guess if your computer sucks at displaying it, then not much point experimenting with capture.

Any tips for improving things at my end?

[BTM_Pix]

Is this in Vimeo or did you download the file ?

[kye]

36 minutes ago, BTM_Pix said:

Is this in Vimeo or did you download the file ?

Both.

Tried VLC and Quicktime on the file.

I even tried to load it into Resolve to convert it to a different format but Resolve doesn't stoop to such lowly formats as mp4 🙂 

[BTM_Pix]

If you have Handbrake you can just do a quick transcode into something you know your computer deals with OK.

Bizzarely enough, as it Vimeo is usually terrible for me, it streams OK for me.

This test by Ed David is going for a similar (minus film) comparison about motion cadence amongst other things so might be worth a look for you 

 

[Towd]

3 hours ago, BTM_Pix said:

On a more serious note, this is a decent thread with some informed discussion about it and also contains a link to a side by side video between real film, Alexa, Red and the Sony F65.

https://cml.news/g/cml-raw-log-hdr/topic/16427778?p=Created,,,20,1,0,0::recentpostdate%2Fsticky,,,20,2,0,16427778

This was an interesting read.

For Kye, watching on a good display is definitely going to affect your perception of whatever you are viewing.  If you are testing 24 fps material, I think reviewing your work on a monitor set to 72hz or 120hz would be ideal.  You may be able to go into your graphics card or display settings an make an adjustment there.

As my 2 cents, I'd just suggest trying to get really granular with measuring individual issues that could affect the overall motion cadence.   From the thread above they go into a lot of issues regarding motion blur and rolling shutter, but I think the good thing is a lot of that could be checked by doing measurements on individual frames to see if dynamic range, exposure, codecs, etc. are having an effect on the visible motion blur.

[kye]

Love that video by Ed David - the commentary was hilarious!  Nice to see humility and honesty at the forefront.

So far, my thoughts are that there might be these components to motion cadence:

• variation in timing between exposures (ie, every frame in 25p should be 40ms apart, but maybe there are variations above and below) - technical phrase for this is "jitter"
• rolling shutter, where if an object moves up and down within the frame it would appear to be going slower / faster than other objects

Things that can impact that appear to be a laundry-list of factors, but seem to fall into one of three categories:

• technical aspects that bake-in un-even movement into the files during capture
• technical aspects that can add un-even movement during playback
• perceptual factors that may make the baked-in issues and playback issues more or less visible upon viewing footage

My issue appears to be that the playback issues are obscuring the issues added at capture.

As much as I'd love to do an exhaustive analysis of this stuff, realistically this is a selfish exercise for me, as I want to 1) learn what this is and how to see it, and 2) test my cameras and learn how to use them better.  If I work out that I can't see it, or it's too difficult to make my tech behave then I likely won't care about the other stuff because I can't see it 🙂 

First steps are to analyse what jitter might be included in various footage, and to have a play with my hardware.

[kye]

OK, analysis of this first video:  (thanks to Valvula Films for sharing this footage).

In a sense, this video isn't well suited to an objective jitter test as the focus is pulled during the pan so everything is blurred for most of the pan.  Regardless, the testing methodology was to create an overlay box and offset the underlying video, frame by frame, to the overlay box, and record the offset.  Like this:

I compared the pan that went to the chart vs the pan away from the chart and chose the one with the greater number of frames.

For each camera I chose a range of frames between when the test chart became too blurred, and when the movement became too small for single pixel measurements.
Where there wasn't a perfect whole-pixel offset I chose the closest one.  Where the offset to the left seemed identical to the one on the right I chose the one on the right.

Here are the results:

The first column is the offset of each frame, and the second is the movement between this frame and the last.  The pan was accelerating / decelerating so the speed went up/down.  

My impressions of this are:

• The numbers don't show any jitter
• My impression of which frames were bang-on vs somewhere in-between didn't seem to indicate that there are any big nasties not shown in this data
• These are all high-end cameras so it is feasible that we didn't find any jitter because there isn't any to find
• I could have gone much more in-depth and tried to offset by fractions of a pixel (Resolve will do this) but on a 1080p image any jitter less than a single pixel is probably invisible

What I learned:

• High-end cameras probably don't have much jitter (not really surprising, but let's start from a known position)
• In a test like this, blurring things isn't a good idea, either from a focus pull or from motion blur, and the more frames something moves the more precise a test would be
• A better test would be to shoot where exposure time is very short, there are fine details to track - both from a lens focus perspective as well as simply having details only a few pixels wide

[kye]

I've put the GH5 to the test.

Setup was GH5 in 5K h265 mode (to stress the camera and get the most resolution), Voigtlander 42.5mm 0.95 lens focused at f0.95 then stopped down a couple of stops to sharpen up.  This produced shutter speeds in the 1/10,000s range and shorter.

First test was to pan and track a stationary object, in this case the corner of a bolt in the fence which was a sharp edge with high contrast.  I put a tiny box around it, I think it was about 3-4 pixels wide/tall, and tracked it on a 4K timeline at 300% zoom:

and here's the results:

Observations:

• Test produced good data, with images being sharp even in mid-pan and margin of error was small (only a couple of pixels) compared to large offsets (60-120 pixels)
• Test was hand-held and between myself and the IBIS we did a spectacular job (I think it was all me, but... 😉 )
• As offset was both horizontal and vertical I reached deep into my high-school geometry to calculate the diagonal offset using both dimensions
• There is evidence of 'ringing' in the movement shown in the fluctuations between ~110 and 120
• This ringing may well come from the IBIS mechanism, as ringing is a side-effect of high-frequency feedback loops (of which IBIS is a classic example)

Discussion:

This 10px P-P of jitter is there in the footage, so the question is if it would be visible under more normal circumstances.  

Let's start with motion blur.  If I had shot this with a 180 degree shutter then the blur would be approx 60px long, making the jitter a 16% variation of the blur distance, which is small but isn't nothing.  Also, if the shutter operated like a square wave with each pixel going from not being exposed to being fully exposed instantaneously then the edges of the blur would be sharp, although much lower in contrast.

What about timing?  10px out of 115 pixels is 8.7%, which if this jitter came from the timing of the frames rather than the direction of the camera then it represents a change of about 3.6ms when compared to 24p which has a 41.66ms cadence, so at a given frame it might be ahead by 1.8ms and two frames later be behind by 1.8ms.

Would this be visible?  I don't know.

[kye]

Here's the original 5K h265 file that I analysed above: https://we.tl/t-YgQIWv1Sps

The frames I analysed above were the first pan from left-to-right.  Let me know what you see if you watch it.

Link expires in 7 days, so don't delay!  Limited time offer!!  *ahem*

[heart0less]

Wow, you're quick!
I didn't expect seeing any actual tests conducted this month, let alone the same week of posting.

Even though it's very technical and all those numbers take some time to digest, you manage to show the results in a very clear and interesting way.
Congrats.

Can't wait for more!

[kye]

10 minutes ago, heart0less said:

Wow, you're quick!
I didn't expect seeing any actual tests conducted this month, let alone the same week of posting.

Even though it's very technical and all those numbers take some time to digest, you manage to show the results in a very clear and interesting way.
Congrats.

Can't wait for more!

I'm good at the technical, not so much with the creative.

If you can, please download the clip and see if you can see the jitter.  What is visible is something I need other people to help with.  If there's enough appetite (especially amongst new camera fever season) then I might even generate some A/B tests and see what levels of jitter are visible.

[Beritar]

I have the GH5, A7III and NX1 and I shoot in 4K. I'm very sensitive to motion and I always found the A7III 's motion cadence very very bad. I thought it was because of the high rolling shutter but my NX1 (hacked) has worse RS, and yet the motion cadence is better. Among them, my Gh5 has the better motion cadence (even with IBIS off), frames are still sharp even while panning. 
 

[heart0less]

Downloaded and watched it thoroughly a couple of times.

And I definitely see some serious jittering.

Though I can't rule out the fact that my PC simply isn't able to smoothly play H265 5K footage.
I know for sure that I can't watch footage coming out from my X-T3 in real time and that's H265 4K.

 

That's the biggest problem with the so-called motion cadence - there are so many variables at play here that it's difficult to point a definite culprit.

 

But I agree with @Beritar.
No matter what I did with my previous Sony cameras (a6300, a7 III), the motion always looked off. Panasonic G85 was much better in this regard.

[kye]

APOLOGIES ALL.  The test above was GH5 4K 150Mbps h264, not the 5K mode.

I just shot a few other tests in other modes and came back in Resolve to look at the files and saw 3840x2160 next to the file I used for the above.

Good news is that I now have 1080p ALL-I, 1080p Long-GOP, and 4K ALL-I clips to analyse, so we'll see if there are differences.

[Beritar]

Like @heart0less I can see jittering on my 120hz screen.

Motion interpolation on TVs is maybe a good way to test motion cadence as well. Bad motion cadence results in a lot more ghosting artefacts and throbbing using the motion interpolation. A7III's videos look ugly with it while GH5's videos are better (again even with IBIS off).

[kye]

I just made this clip in Resolve using a few keyframes.  Not only will it have zero jitter, but it will also have no RS.

Use it as a test for playback jitter.

https://www.sugarsync.com/pf/D8480669_08693060_8819366

[kye]

OK, here's all four combinations.

The blue line is the movement per frame.  The orange line is a trend line (6th order polynomial) to compare the data to and see if the data goes above then below then above which would indicate jitter.

Looking at it, there is some evidence that all of them have some jitter, possibly ringing from the IBIS.  I got more enthusiastic with my pans in the latter tests so there's less data points, so they're not really comparable directly, but should give some idea.

They appear to be similar to me, in terms of the up/down being about the same percentage-wise as the motion.