A short stabilization demo

[Perrone Ford]

I am currently doing some post work for a small indie film, and several scenes needed to be stabilized. The director was adamant on set about not wanting the feel of a tripod and banned them from the production. However, a few shots came out a bit too rough.

I put together this demo today to show what post stabilization could look like while keeping a somewhat organic and non-static feel...

Particulars:

Shoot offhand
Canon 5D
Lens is unknown

Software: Virtualdub
Filter: DeShaker v2.4
Params: Available upon request
(this filter is also notable because it deskews cameras with rolling shutters.)

YouTube - Stabilization Demo

[Marcus Martell]

HI PERRONE, do you think this stabilyzer would work with helicopter shaky footage?

Thx man

What kind of movie are u shooting?

[Perrone Ford]

I have never seen raw footage from a copter. If you have some, I could give it a try.

I got hired on in October to AD a twilight zone type of movie, but we lost the gaffer the week before production. So I filled in for him, did a little DP work, and tried to maintain my AD responsibilities. Let's just say I was BUSY!

Now, I am acting as colorist in post, and doing some cleanup of the footage as we go along (like stabilizing some of the handheld work).

It's a challenge, that's for sure. But I'm enjoying it tremendously.

[Steve Kalle]

So, does VD output a reduced resolution file to compensate?

I am a bit familiar with VirtualDub but never thought to use it for pro work. Does it support XDCAM EX?

How long does it take to render using the DeShaker?

Thanks

[Perrone Ford]

No, it does not output a reduced resolution file. However, since I knew I'd be doing the post, I asked for a 10% overscan when we were handheld so I'd have room to maneuver in post. Virtualdub has numerous filters that are up to spec for pro-work, and some things work better than my pro apps, including pulldown removal, resizing of footage (it has Lanczos resampling which is better than any NLE has natively), denoising is on par with NEAT Video but free, and the deskew/deshaker program is better than anything I've seen in any NLE. I've seen what AE has, and I think this is better.

Virtualdub supports .AVI files ONLY. Input and output. There is a hack to support .R3D files, but only with limited color space.

Deshaker takes a LOOOOOOONG time. There are numerous parameters to tweak the output and some of them make things really slow. Whereas most deshaker programs allow you to pick a couple of points for reference, I am able to pick as many as I like in a grid pattern. If I want pro level results, I do my 1920x1080 scan in a 10x10 pixel grid. Believe me, it's REALLY accurate.

[Marcus Martell]

Do you use this deshaker with Vegas?How much does it cost?

thx a lot

[Perrone Ford]

Vegas has nothing to do with the Deshaker. The program I use is called VirtualDub. Many people make "filters" for VirtualDub. Think of them like plug-ins for Vegas or Final Cut. Except, Virtualdub is free, and most of the plug-ins are free also.

[Marcus Martell]

Thx Perrone, so does Virtualdub handle HDV (m2t) files?
Is this deshaker free also?

thx a lot Perrone

[Perrone Ford]

Yes, Deshaker is free. No Virtualdub does not ingest mpeg files. It's .AVI or nothing.

[Chris Jeremy]

Originally Posted by Perrone Ford

No Virtualdub does not ingest mpeg files.

You can get an MPEG-2 plugin for Virtualdub at MPEG-2 plugin for VirtualDub

[Perrone Ford]

You know I forget about that thing. Thanks for reminding me... Not that I plan on bringing in any mpeg2 files, but it's good to know anyway. Actually, I wonder if it can read native XDCamEX files... THAT might be worth some investigation!

[Update]

Nope, can't read XDCamEX... would have been SUCH a win.

[Chris Barcellos]

Perrone:

I have done some DeShaker with my 5D footage, but would love more info on the parameters you are using to get there with this DSLR footage.

Can you post some screen grabs of the setting screen ?

[Perrone Ford]

Easier said than done. While Pass 1 params are fairly fixed, pass 2 is where the meat of the matter comes in. And those setting change drastically depending on what you are trying to do with the footage.

So for Pass 1 this is what I do:

Video Output: Motion Vectors

Block size: 10 pixels
Scale: Full
Use Pixels: All
Color: RGB

Everything else is default except the last option which I change from from 15% to between 3 and 5% depending on the video. In a very dark scene where there is detail in the shadows just uncheck it.

On Pass 2 I set my edge compensation to none. I do this because I want to control the framing back in the NLE. The software will center the image for me, but I get to control where I want to position my frame, and the choices I make might be different than what the software wants to do.

If I am correcting VERY small movements, I will use the previous frame info to fill in the black frames. Usually I don't do this because I want them to be black so its easier to correct in the NLE

Extra Zoom factor I leave at 1. I don't want the program zooming in my frame for me.

The Motion Smoothness is where things get interesting. The larger the numbers, the more smooth the video. I generally use smoothness setting around 1000 for all but zoom, which is kept a 1. If you zoom during your shooting, this would have to be changed.

The max correction limit is where you make tradeoffs. The larger the numbers, the smoother your video will be. But if there are large corrections (like you fell down while filming) then your screen will go black while it tries to compensate for that radical motion. Small numbers tend to only allow small corrections.

There are three axis represented here. Actually 4 if you count the zoom. Since I don't zoom, I only deal with three. And you need to look at how your unstability comes about. If you are pushing a wagon down the street with the camera on it, changes are you aren't going to need much horizontal correction. But you will need quite a bit of vertical, and maybe a hair of rotational. If you are holding the camera in your palm, you will need correction in all 3 axis.

The first pass takes a LONG time. Especially if you change the block size to 10 like I do. I find that doing that allows for much finer control and gives less jello effect to the video.

The stabilizer works best of small organic movements. Quite like a steadicam. Heavy footfalls will jar the video and be hard to correct. Swimming motions are also problematic. But overall, the program does a marvelous job. Stabilizing something with a soft background REALLY looks filmic. The more detail you hold onto in the background, the more it tells on you.

[Chris Barcellos]

Perrone:

Thanks for the in depth explanation. That gives me a feel of where to work on things. I was definitely dealing with some jelloish results....

[Perrone Ford]

One other thing. Over on the left where it asks if your camera has a rolling shutter. It asks you to put in the amount of correction. I experimented a while with this and came to conclude that 68% looks very nice for the EX1 and 75% looks nice on the 5D. You'll have to play around a bit to find good settings if you have a different camera.