Zooming in Stereo 3D anyone?

[Clyde DeSouza]

This probably would be of interest to all the 4K and 5K 3D shooters out there.
I know it's ken burns like...but let's give it a proper S3D christening shall we?. I call it the Clydozoom.

here's a brief intro...
In Stereoscopic 3D Cinematography, doing a zoom will “compress” or Flatten depth. This is due to the nature of telephoto (zoom lenses). You normally “dolly in” on a subject of interest rather than zoom-in in 3D, so as to avoid that layered or flat 3D effect also called “cardboarding”.
So called, because even though the scene looks 3D, it’s as if every layer of depth has been hand cut and placed into the scene.

Performing a zoom in 3D… I call it the “Clydozoom”, is akin to the Ken Burns effect, but with the specific goal of executing a Zoom rather than a pan and scan or dolly-in or other such effect, and is aimed solely for Directors or Cinematographers who for any specific reason need a Zoom effect in their Stereoscopic 3D movie.

The Zoom needs the following to be true in order to execute successfully... read the rest here: http://realvision.ae/blog/2011/04/stereoscopic-3d-cinematography-zooming-in-3d/

 

[Aaron McLane]

Great post! We recently used that technique on a Redbull cinema commercial we did with Brainfarm. it's important to note you are changing the effective IOD when zooming in or cropping a stereo image and even with HIT fixing you can only go so far especially if the IOD was maxed to begin with. In your example you can see the forced perspective effect happening during the digital zoom which is very realistic and of course impossible when doing a Ken Burns push on a 2d image.

 

[Clyde DeSouza]

>> it's important to note you are changing the effective IOD when zooming in or cropping a stereo image

Hi Aaron,
Thanks for the kind words. Regarding the above line... here's my reasoning

1) If you have a 4k image (stereoscopic) and you scale it down to HD (for final projection), there would be no change of IOD or I/A.
2) This is exactly what the technique describes. Capturing an overscan image such as 4k or 5k knowing full well that final display medium/size will be HD or 2K.
3) At no point do we scale this 4K source image to larger than 4K size. The "zoom" is from always starting with the scaled down version and using the extra rez available in a4K image to scale up to the initial 4K only, never more.

Cropping a perfectly captured 4K stereo image would not change any effective IOD either. It's just like a "window" (in this case an HD sized window) panning across the canvas that is the 4K image. If you were happy with the IOD that was set when the 4K image was captured, cropping any section out of this image would not alter the IOD in anyway.

This is how I see it... yet of course, i'm willing to learn more and be corrected.

Best Regards
Clyde

(P.s any chance of seeing the RedBull clip?)

 

[Doug Horton]

I'm on dinosaur speed wifi so I can't watch the youtube clip at the moment, but I'm trying to wrap my head around the concept mentioned just above.

You set an IO based on the constant factors (sensor, lens, FG object, BG object), so if you're windowing the sensor (more accurately the image), are you changing any of these parameters? Not exactly, however by windowing the sensor, you are changing the FOV, just like if you were shooting on a R1 in 2k, you're not really using a 35mm lens at that point, so does this same principle apply here? True you're pulling out to your original 4k frame (the one you calculated your shot for), but typically we would do an I/O pull in this case.

True, there is depth loss in 3D zooms, but, like anything else in this industry, one can use this flattening effect to their advantage if need be. I'll watch the video tonight and have a better idea, but either way I'm excited to see!

 

[Clyde DeSouza]

Hi Doug,
Again let me repeat the steps and see if i'm missing something here...

1) Assuming you shoot a RED 4K stereoscopic image and want to make a final out for HD only? what do you do? You woul'nt pull interaxial right? You'd simply down rez the image to 1920 x 1080

This is the basis of what we are doing here. We start off the zoom with resizing the original 4k or 5k image down to the lesser "final out" (HD or 2k Stereo).

2) Now building from here, we "window" the image..i.e start re-framing (using an HD window or 2k window) to the subject of interest in your 4k image to perform the "zoom".
At this point your still working from a 4K buffer. and you never scale up more than the size of the 4k original in order to achieve the zoom.

Regards
Clyde

 

[Aaron McLane]

Hi Clyde,

I'm wondering what you mean by making your final for HD only. My understanding is that with 3D your calculations are based on viewing size not pixel count, meaning that if you are shooting for a final viewing on a 65" HD 1920x1080 monitor your math would be compleatly different then if your final viewing was on a HD 1920x1080 projector on a 40' screen.

Also if you crop an image or switch a Red to 3k from 4K you have to recalculate based on a smaller sensor size. You can get away with most of it on monitors but when you go into a 40" screen not all of the tricks work the way you want them to. We've learned that the hard way when someone in post cropped or pushed in without considering the math.

The RedBull spot we did will be shown in 3D in front of Pirates 3D and Transformers 3D this summer and can be seen in 2d on TV and the web. The Stereographer on it was our in house physicist Leonard Coster directed by Curt Morgan of Brain Farm

 

[Clyde DeSouza]

Let me clarify...
a) let's assume your shooting for Cinema, and overall depth budget is 2% (for HD final out)

b) Your NOT cropping or switching at acquisition time with the Sensor.. This is a Post Production zoom technique. Your acquisition is always at 4K or above

c) if your 2% overall depth budget is all sorted and your 3D image looks good at 4K, you down rez the image for HD final out (regarless if for cinema or 3DTV. Yes your 3DTV version will not look as great due to budget being set for Cinema)

d) Again just to re-iterate .. Your NOT framing/cropping the Sensor. Your doing all this in post from the original 4K pixel resolution data that you have. and you Do not scale above 100% (of the original 4K image).

Think of it as if in a 4k image you have 4pixels to represent a thin vertical line.,,,, in a down rezzed image to 2k, you will have 2 pixels to represent that same vertical thin line.
The common denominator for projection screen is still the Cinema 40foot screen.

if your parallax is legal to begin with (for the 40 foot screen and for the 4k image), downrezzing the 4K image should not squish the parallax or, in other words, the distance between left /right images.


The overall depth budget should be set keeping HD or 2K final out in mind, and capture should be done at 4K or above.
This is how I see it, hopefully i'm explaining it proper.

Math is good, but 3D should not be have a "paint-by-numbers" approach,,, I always saY
yes...and neither should it rip eyeballs out with extreme background divergence.

 

[Aaron McLane]

Here's that Red Bull spot.... can you spot the ClydoZoom?

http://www.e3dcreative.com/blognews/

 

[Clyde DeSouza]

The ClydoZoom explained further...

I got a lot of Cinematographers and a few Stereographers as well asking questions, and as we are all learning (including myself) I thought it would be best to explain with pictures. I'm open to further insights...

The Setup: We will be using an After effects composition of HD size and an HD image (1920 x 1080) to represent a 4K master image. In reality if we shot 4K originally, we would setup an Aftereffects composition of HD (or 2k) for our final out, and the original 4K master image in 3D would be scaled down to fit this final out composition.

...for our demo, as we do not have a 4K stereo image to play with, I use an HD 3d image to pose as the 4K master image, and I will setup an after effects composition of 640 x 360 pixels (3 times smaller than HD) to represent the final out HD screen size. All final out is meant for Cinema Screen Projection.

One of the biggest concerns I got via emails, was from Stereographers and Cinematographers about Parallax magnification if we "zoomed in". Another concern was FOV changes, if we crop the sensor / image.

Figure 1: The Original HD image of 1920 x 1080 (meant to simulate the "4K" master image)

In Figure 1 we see the original HD image (simulating a 4k image). The onscreen ruler (excellent freeware) shows us a pixel disparity of approx 18 pixels. (for the Purists out there: ignore the fact that there is vertical mismatch of a few pixels in this image, and further ignore the fact that 18 pixels is too much for the scene, this is an uncorrected, and un- HIT image....) Let's assume we are happy with this depth budget. Now... if we were happy with this 4K stereo image and 18 pixel disparity between left/right images, the first step we would do (regardless of zooming etc) would be to scale this down to final out of 1920 x 1080 if our final out delivery platform was HD.

This brings us to figure 2 below

figure 2: the final out comp of 640x360 (meant to simulate an HD or 2K final out)

Here we setup a new composition setting of 640 x 360 pixels (simulating an HD final out). So we thus scale down our original "4k" image to fit into this final out for HD projection. The pixel disparity is now 8 pixels above, and is actually smaller than the original 18 pixels @ HD resolution.

So if this were an actual 4K image and the pixel disparity of the 4K image was 18 pixels, it would be reduced by almost the same factor when scaled down to HD rez.

Still with me?....

Figure 3: At 95% scale, we are back to a 17 pixel disparity (well within our original image's disparity range)

Now... we perform the ClydoZoom, which is a combination of "windowing" or panning, or re-framing the oversized image along with uniform scaling of it, using all of the available resolution data of the original "4k" image at our disposal, so long as we DO NOT scale ABOVE 100% or the original image.

Thus at even 95% scale up (or...zooming in), we are still back to a 17 pixel disparity... as we remember, right at the beginning we were "happy" with our 18 pixel parallax to begin with!

I hope this clarifies the ClydoZoom in more detail.

Lets address the other concerns:

Cropping the Sensor: We are not cropping the sensor at all. This is a post production technique, so all cropping is done on the master 4K image. Next in an Optical Zoom, arent we always changing the FOV? That's what a zoom is really, a close up, and thus a change of the Field of view in the simplest sense.

Again... (Note: These workflows are still experimental and being worked on at Real Vision. The techniques have not yet been thoroughly tested for every kind of scenario in 3D. Readers thoughts and further ideas are valued and welcome)

Original post here: http://realvision.ae/blog/2011/04/stereoscopic-3d-cinematography-zooming-in-3d/