How you do this
 

Not a simple vanishing point :)
http://red.cachefly.net/Glidetime.mov

Re: How you do this
 

Good question - one possibility that comes to mind would be to model and pose the characters in a 3D app or model in something like Poser and insert into a 3D scene and then animate the virtual camera. I didn't see anything that couldn't have been done that way. The way the shadows of the shooters are projected on the wall towards the end makes me think this was how it was done - also the face of the first shooter looks as though it could have been a modeled figure. The other characters go by quickly enough that it's hard to tell for sure.

Might be all wrong (as usual) but just a guess.

Another thought would be to pose live actors using wires to support them in their poses and hang the hats etc the same way and roto out the wires and track the scene and add the splatters and flashes in a 3D app. But I think it would have been hard for the characters to stay still enough long enough even with wires supporting them. I've looked at the thing several times and I don't see even the slight motion I'd expect if they were live.

Regardless of how they did it it's pretty effective.

Edit - even though it's physically incorrect. The victims were actually lined up against the wall of the garage before being shot - if you google around enough you can find photos takes at the scene showing bodies up against the wall.

Re: How you do this
 

Here's another guess:
Assuming the actors are real, then the scene (actors plus green screen plus set) is the result of a bullet-time effect in which the cameras were 'fired' at the same time, with interpolated scenes created in between via retiming.

The extended set was created and added via a 3D camera tracker. The camera shake appears unnatural. They have taken great pains to avoid the path the camera moves through, which wouldn't be an issue in a full CGI scene.

Re: How you do this
 

Interesting thought - could be, although if you had multiple cams scattered about it would be hard to keep them out of view of the other cams - but then again, the cameras could all be facing the same way and located together. I was a bit puzzled by the fact that all the shots were facing pretty much the same direction. It was as if the camera path was studiously chosen to avoid seeing anything to the opposite direction.

I was also wondering about a lot of the sort of reflections that showed up as the camera panned.

I'm pretty sure though that a lot of the stuff like flashes and cigarettes and hats floating in the air were added into a tracked shot.

Another idea would be that there were only two or three actual high res shots and all the panning was done within a higher res image.

But the changing perspective during the pan kind of made me rule that out.

Probably what they really did was to flash freeze human actors and hang them up and film quickly before they started to melt. (or mannequins???)

This stuff is fun!

Re: How you do this
 

Do you think this all may CG? Have talent stand in then take a bunch of pictures so the modeler has something to work off of?

Re: How you do this
 

Actually I'm inclined to think so. Posing the characters is nos so bad - something like Poser makes it quite easy and high end packages have more than adequate tools for hair and cloth - photo projection on models is also available in some if not all the major packages. C4D certainly has a lot of texturing and projection tools available.