Time Slice

[bartrobinson]

I've heard this is done through stereophotogrammetry using a stereo pair (two images taken at slightly different angles) of reference images. Well, yes and no. It's not exactly stereo or photogrammetry (at least that's not what most of what you see is and it's not how it started). What it is: an array of cameras (photo, film, or video) synchronized to take pictures in unison from slightly differing angles or positions. In post, you then load all of the sequences. From there you have a number of directions you can go. You can do "frozen time", by choosing the same time frame from each camera and playing them one after the other. To get rid of the jerkiness you can use a retimer like RE:vision Effects Twixtor, Realviz ReTimer, or discreet Inferno to get inbetween frames. You can do motion effects by simply choosing frames from each camera at increasing or decreasing times then retime those frames. That will let you use slow motion forward or reverse or mixed motions. Note that these techniques rely on relatively little in the way of cg and compositing. You basically shoot it with your huge camera rig, edit and retime it in your editing system.

Now some of this could be achieved by photogrammetry with plenty of computer and compositing work and could also probably be done in a number of ways. Shoot motion control first with subject then without. Pull subject into 3d with photogrammetry software. Animate and composite with clean plate. Edit with original sequence. OR Shoot subject motion control on bluescreen, shoot background plates, retime subject, composite. The point is, there's about 10 different ways to do things. You just need to be able to find the best solution for the situation.


What kind of software is available for stereophotgrammetry? Is it just regular image modeling programs (like Realviz's ImageModeler or UZR 3d?) Or am I going to need something more advance?

Yes there's Realviz ImageModeler. There's also EOS Systems PhotoModeler and I'm sure there's others.

[Aneks]

I was both intruiged and amused to hear John desJadin say that after all the hoo-haw around 'bullet time' in matrix one they juut threw all the multicam stuff away in 2 and 3 and ended up making digital environments precislely because it was so unpredictable, combersome and unsatisfactory a technique.

the clip Gondry did for the Stones seems to be a much simpler technique than a multicam 'timeslice'. Don't have the DVD with me right now but i seem to remeber it looking more like it was a series of morphs between stills and short image sequence taken from two or three cameras mounted on the same rig !

In terms of stereo photogarmmetry have a read of cinfex and the siggraph white papers on the universal capture (Ucap) and enviromental capture systems on the matrix. They are the extreme version of these techniques.

A friend of mine is busy designing a similar type of application using 3dEqaulizer and its ability to calculate geometry and motion from stereo camera setups. Also have a look a the pixel farm suite of tools these look like they could be the best comercially available solution. From memory these all use optical flow technology as the basis of the image analysis algorithm

[cache]

hi!

I'm really interested in how the morphs were done in the rollings stones video, but i could not find anything useful with google, so i thought i ask from You !

is it some kind of automatic morphing; or was it done manually ? (that sure would be a lot of zombie work to do morphing a videoclip length footage frame by frame... :eek:

please help me out !
thanks for the answer !

[Dutchman]

Really interesting links & information in this thread! I think I can also contribute a bit to it :) :

‘Time Stands Still’ Burton encouraged Mack to be as creative as possible within a modest visual effects budget to achieve suitably ‘tall’ effects for Big Fish. For instance, when working on the scene where time stands still, and Edward spies Sandra, his future wife, Mack considered using numerous techniques to achieve the effect. In the end the solution he chose was a blend of practical cinematography and digital effects.

“We considered using camera arrays for this scene, but that wouldn’t work,” explains Mack, “because you can only do that if the cameras aren’t looking at the path they follow. We had a bunch of items including popcorn in midair, juggling pins, an acrobatic cat halfway through a high fall, and a flaming fire log being juggled on a woman’s feet. Those were all created as computer-generated 3D elements and added to the scene. But for the people we used real actors and got them to hold as still as possible. We even sent them to a mime school for a day to teach them how to hold still, but we still had them blinking and swaying and jiggling because keeping completely still is very difficult to do. We carefully cut out patches and manually tracked and warped them to fix the slight movements. If, for instance, the clown’s arm was moving too much, we’d just cut out his arm. We would use this one frame of his arm and track and warp it over the frames of accidental movement. We would then morph from the tracked bits back to the real arm.”

This scene was the most difficult for Mack with regard to developing the technical solution. The hard work paid off, however, with the result being a very seamless and realistic scene where time stands still for everyone but Ed as he edges his way towards Sandra, dodging the stationary tumblers, hoops and falling cats, and brushing aside popcorn poised in mid air.
[From CGNetworks.com production coverage (http://www.cgnetworks.com/story_custom.php?story_id=2012&page=2) :) ]


So also over in the Big Fish case (one of my favorite movies :) ), with a very amazing time-slice shot (of about 40 seconds!), they just used people standing still & the use of some rotoscoping/compositing techniques! :twisted:


To Cache: No, I don't think that its manually done in the way you think: for morphing you have applications that automatically in-between the morphs... (if that's what you mean ;) ) For example the cheap Morphman (http://www.stoik.com/morphman/) can do this trick for you! :)


If you want to use this effect in a feature you work on, I think there are 2 usefull methods:
1) Just taking some photographs, and 'undress' them in Photoshop (so get a cleanplate, and a character plate, and a backgroundplate and so), and combine them in 3D in for example After Effects! In that way (ánd with some simple reshape-tricks and rotoscope tricks and so) you can come very far! :)
2) Just take some footage (moving), but with the people standing as still as possible (> Big Fish method), maybe with help of wires and so, and shoot a cleanplate, and later use rotoscope-tricks and so to make it PERFECT! :)

I hope you friends understand me :) & Good luck with your projects, especially if you're about to use time-slice! :buttrock:

Greetz,
Gijs

[pierrejasmin]

hi!

I'm really interested in how the morphs were done in the rollings stones video, but i could not find anything useful with google, so i thought i ask from You !

is it some kind of automatic morphing; or was it done manually ? (that sure would be a lot of zombie work to do morphing a videoclip length footage frame by frame... :eek:

please help me out !
thanks for the answer !


This is pretty old now. Buf then used a mixture of motion estimation (like Twixtor does) and Morphing (like RE:Flex does)...

I have done some cool stuff with 2 camera views and RE:Flex Motion Morphing (which has motion estimation inside), assuming one view is morphable to the other. (the simplest is shoot stereo if you have a way to collect 2 frames at the same time - as opposed to a field based swap view process).

Basically with RE:Flex Motion Morph:
1) I set From Source to one camera and To Source to the other
2) I establish key-correspondance manually with small splines
3) turn on auto-align
4) Then you can simply animate the Global Warping and Blending from one to the other and sometimes you have lattitude for Warping to go below 0 and over 1.0 to extrapolate so give you more POV change lattitude. (as in 0.5 is the view in-between).
With a bit of editing on your source clips (defining some specific key-points and switching source at these frames) you could expand that to more cameras.
If you have an actor in the foreground far away from the background, it's better to have the actor isolated and do a two pass process.
5) I also did play with sequencing stereo stills with RE:Flex (for example: http://www.revisionfx.com/rflx/oaken1.mov ) -- This is done with 9 stereo stills. One thing to check out, in this "more fluid" effect, I was missing a view (one film roll was damaged) so I ended up with the console on one side in one frame and on the other side in the next frame and the actor in both case in the center of the frame... so there is a bit of wobbling in the center -- so bottom line, careful planning is important.



Pierre
www.revisionfx.com (http://www.revisionfx.com)

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.