PDA

View Full Version : Motion smoothing (request for footage)



stephenlnoe
08-26-2005, 09:01 AM
Hi,

I'd like to get some footage (if you get the time) of the motion smoothing function of the HD100. What I'm asking is that you set the camera on a tripod and then do two shots, one with smoothing, one without.

The shot I'd like is a hand/arm waving in front of the camera (fairly close range). I want to see the blur or stutter (whichever the case may be) for a simple motion. So in essence a steady background and the motion of an arm/hand waiving.

If there already is such footage (m2t files) out there please point me in the right direction.

Nate Weaver
08-26-2005, 09:25 AM
On this page:

http://www.gpjco.tv/~dvdoctor/

There's two clips of cars passing on a road just like you want. Locked down, no movement except the cars.

Elsewhere there's discussion of the two clips with stills. The current theory of motion smoothing based on these two clips is that the MPEG encoder just superimposes two frames over eachother for the motion smoothing effect.. There's some debate as to whether the two frames in question are captured 1/24th apart or 1/60th (if the latter, that would have implications for how the camera section of the HD100 works). I myself think it's just two regular 1/24th interval frames (shutter speed being at whatever you set).

Also, about the clips on that page, the author later mentioned that the shutter speed on those clips was 1/500th! Thats why the non-motion-smoothed one is strobing.

dashwood
08-26-2005, 09:55 AM
Hi,

I'd like to get some footage (if you get the time) of the motion smoothing function of the HD100. What I'm asking is that you set the camera on a tripod and then do two shots, one with smoothing, one without.

The shot I'd like is a hand/arm waving in front of the camera (fairly close range). I want to see the blur or stutter (whichever the case may be) for a simple motion. So in essence a steady background and the motion of an arm/hand waiving.

I just shot exactly that test.

Camera settings: 720p24, shutter off (1/48th), ND #2, 5.6~8, focal length 5.5mm, focus set at 4 feet.

Right-click and download these m2t files to your computer.
Motion smoothing off (9.7Mb):
http://homepage.mac.com/timdashwood/.Movies/motion_smooth_off.m2t

Motion smoothing on (11.4Mb):
http://homepage.mac.com/timdashwood/.Movies/motion_smooth_on.m2t

Personally, I would leave it off if you want a true film look.

This is actually a pretty good test for judging compression artifacts on leaves moving in the breeze.

Tim

dashwood
08-26-2005, 10:03 AM
Upon closer inspection of my own test I would definitely recommend leaving motion smoothing turned off.

As you go through frame by frame on the clip with motion smoothing on, you can sometimes see residual MPEG compression artificacts in the pattern of where the hand was in the previous frame.
This didn't happen when it was turned off.

As a side note, the other interesting thing is that 1/2 second into the "motion smoothing off" clip you can see a yellow jacket fly from the right to the left and his motion blur stretches about 1 foot on each frame. I'm surprised the codec captured this small detail so well.

Tim

Nate Weaver
08-26-2005, 10:11 AM
As you go through frame by frame on the clip with motion smoothing on, you can sometimes see residual MPEG compression artificacts in the pattern of where the hand was in the previous frame.
This didn't happen when it was turned off.

The other interesting thing is that 1/2 second into the "motion smoothing off" clip you can see a yellow jacket fly from the right to the left and his motion blur stretches about 1 foot on each frame. I'm surprised the codec captured this small detail so well.


You're describing exactly what happens in the clips I linked, except the "MPEG artifact" IS the effect. Since the guy shot his test with a fast shutter speed, motion blur from the camera head isn't such a player.

Motion smoothing seems to be a little crude, at best. If it's something they incorporated in the MPEG encoder, which it seems, then it IS a bit clever though.

dashwood
08-26-2005, 11:10 AM
Motion smoothing seems to be a little crude, at best. If it's something they incorporated in the MPEG encoder, which it seems, then it IS a bit clever though.

Maybe it's an example of creative marketing: calling a inherit defect in MPEG a "feature!"

stephenlnoe
08-26-2005, 07:11 PM
I just shot exactly that test.

Camera settings: 720p24, shutter off (1/48th), ND #2, 5.6~8, focal length 5.5mm, focus set at 4 feet.

Right-click and download these m2t files to your computer.
Motion smoothing off (9.7Mb):
http://homepage.mac.com/timdashwood/.Movies/motion_smooth_off.m2t

Motion smoothing on (11.4Mb):
http://homepage.mac.com/timdashwood/.Movies/motion_smooth_on.m2t

Personally, I would leave it off if you want a true film look.

This is actually a pretty good test for judging compression artifacts on leaves moving in the breeze.

Tim

Some artifacts it's true but all in all, for that much motion it is very good. I zoomed into an insane value (1600X) and it holds really well frame by frame in both the smooth motion and non-smooth motion. Sony should eat their hearts out. Also the CX color corrector did not squak at the color gamut. It was completely legal. It's really good.

I think to pair this camera with a dynamite lens is a powerful combo. I also think for the format and storage requirements (tape or an onboard hard drive), this is a terrific option for documentary videographers.

Now for an acid test (real world). Has anyone tried this with an HDTV (flat) via component out of the camera? I'm talking from the tape, not a live feed.

Here are the screen shot's zoomed (@ 1000X)

Smooth Off (http://home.comcast.net/~stephenlnoe/smoothoff.png)

Smooth On (http://home.comcast.net/~stephenlnoe/smoothon.png)

Either way you're getting some blocks which is only natural, it is DV. You have to zoom mighty tight to pick up macroblocks. I think they've done a good job with the codec.

For all the engineers in the house click here (http://home.comcast.net/~stephenlnoe/subpixel.png)for the subpixel routine I used on the zoom just in case you want to try to get the same results in your NLE.

overall it is good...

mezelf27
08-26-2005, 08:12 PM
macroblocks zoomed that far in? ... microblock is what you saying then :-)

dashwood
08-26-2005, 10:18 PM
Now for an acid test (real world). Has anyone tried this with an HDTV (flat) via component out of the camera? I'm talking from the tape, not a live feed.

I'm trying it on Monday with a DecklinkHD. We'll try digitizing 10-bit uncompressed off the component out and see what happens.

Tim

stephenlnoe
08-27-2005, 02:19 AM
macroblocks zoomed that far in? ... microblock is what you saying then :-)

There are pixels (1x1), blocks (8x8), and macroblocks (16x16) (click here for screenshot) (http://home.comcast.net/~stephenlnoe/macroblock.png). The macroblock (definition) is the fundamental unit for motion compensation and will have motion vector(s) associated with it if is predictively coded. In this case the prediction 'method' is either the HDV codecs' built in motion prediction method or JVC's "smooth motion" method.

This makes me wonder what the intent was for 'smooth motion'. For instance, in post production FX, Lanczos is generally good for most general use (small scaledown and ups) it is well known for its edge-enhancement but some macroblocks might be still visable. For extensive Zoom or scaling I would recommend Cubic B Spline. I'm pointing this out because most algorithmic methods are created for a specific purpose.

It makes me wonder what the specific purpose "smooth motion" was developed for and under what circumstances to use it to get best results (as JVC intended).

Does anyone know an insider at JVC engineering to find out what the intent is for "smooth motion"?

MattDavis
08-27-2005, 09:56 AM
It makes me wonder what the specific purpose "smooth motion" was developed for and under what circumstances to use it to get best results (as JVC intended).

FWIW, I've noticed that on my 1.33 GHz G4 PowerBook, playing m2v clips with VLC, the "Smooth Motion" clips play back better (fewer if any dropped frames), and that the waving arm has better blending of frames. Blending isn't the right word - better progression? Dunno. Nothing scientific about it, no theory, just the way I'm experiencing it.

Maybe it's to help playback?

dashwood
08-27-2005, 01:07 PM
FWIW, I've noticed that on my 1.33 GHz G4 PowerBook, playing m2v clips with VLC, the "Smooth Motion" clips play back better (fewer if any dropped frames), and that the waving arm has better blending of frames. Blending isn't the right word - better progression? Dunno. Nothing scientific about it, no theory, just the way I'm experiencing it.

Maybe it's to help playback?

A 1.33Mhz single processor will definitely drop frames on an MPEG2 transport stream at 720P. Try playing it back with MPEG streamclip from squared5 and you may have better results. It is a free download. You can also convert the streams into any other format.
http://www.alfanet.it/squared5/mpegstreamclip.html

I'm running a dual 1.2Ghz G4 and I drop frames on playback of 720P60 MPEG streams. I'm planning on getting a dual G5 soon.

It is still under 25mb/s like DV, but it is alot harder to decode and present the higher resolution in real time.

Barry_Green
08-27-2005, 02:03 PM
Does anyone know an insider at JVC engineering to find out what the intent is for "smooth motion"?
My understanding of it is that they intended to simulate, as best as possible, the look of 60p video; to give it the "reality" look. At least, that was (AFAIK) the intention; it's certainly not what the result is.

Nate Weaver
08-27-2005, 02:31 PM
My understanding of it is that they intended to simulate, as best as possible, the look of 60p video; to give it the "reality" look. At least, that was (AFAIK) the intention; it's certainly not what the result is.

Really? I always thought it was an attempt to make people happy that would like to shoot in 24, but think it's "strobey".

To which I'd say the answer is to shoot with a 1/24th shutter, but different strokes...

Barry_Green
08-27-2005, 04:06 PM
That's exactly what I'd say too -- shoot with a slower shutter speed.

But you may have a point -- it may be to just minimize strobing. I always thought they were trying to overcome the lack of 60p, but that would only really apply if it only worked in 30p mode -- your point is quite valid in that it works in 24p mode as well.

Either way -- I don't care for it. I'd much rather do as you say, and use a longer shutter speed -- I think the results are a lot more along the lines of what one would expect, vs. the double-image method of Motion Smoothing.

stephenlnoe
08-27-2005, 07:26 PM
I'm trying to ascertain what the intent was from the beginning (as put forth by JVC). In other words, what scenario (acording to JVC) would 'smooth motion' be most optimally used? They've developed the 'smooth motion' method with something in mind. I do not know what that is and the brochure does not give enough info to realize the intent of 'smooth motion'. In JVC's test lab the 'smooth motion' method was developed under what circumstance?

I have only one contact at JVC and he is by snail mail:

Dwayne Kersey
Professional Products Support Specialist/Engineer
JVC Americas Corporation
1700 Valley Road
Wayne, NJ 07470

If anyone else knows someone from JVC, can you drop them a line and ask what the true intent is for 'smooth motion'?