Monday, January 08, 2007

Examples of Automavision and a proposal for Automavision 2.0

Following on from my discussion of Lars Von Trier’s Automavision and Lookey I thought you would like to see some examples of Automavision. David Bordwell, the author of a phenomenal number of outstanding books on the subject of film has written a very interesting blog post on the subject here. He has had the good fortune of viewing The Boss Of It All, unlike myself and capturing some screenshots. The effect is intriguing. Automavision appears to create unmotivated, and classically imperfect framings which Von Trier accentuates by cutting rapidly between very similar shots. Bordwell notes that the result is that almost every cut is a Jump Cut, a violation of the 30 degree rule that causes the image to jump uncomfortably and creates ambiguous temporal relationships between the shots. The same effect occurred in Dancer in the Dark (see a video example here) and can be said to have contributed to the overall discomfort felt by viewers of the film.

What is most interesting about Von Trier’s use of Jump Cuts is that, whilst they abandon the classic continuity style’s preservation of temporal continuity within scenes they still retain a clear cohesion that allows the viewer to understand the action represented. It is almost as if Von Trier, in consciously violating the dimensions of continuity prescribed by the classic continuity style he is revealing extra dimensions of continuity that he uses to create extra significances within his films.

Also, as noted by Antithesis Boy in his comment on my last post the Automavision technique shares a lot of similarities with virtual camera control in videogames. Automavision uses a computer to randomly generate framings for the camera and the result is non-classical framings. Videogames often take place in a 3D virtual environment and require a virtual camera to follow the action within this space. This camera is computer controlled and the objective is to create the best framings possible (e.g. Jhala and Young, 2006) but the result is often unacceptable and bizarre framings. The difference between the two systems is that a Virtual Camera is positioned relative to the objects within a scene where as Automavision (as far as I can make out) does not care what the scene is. This is why Automavision loses significant objects off the edge of the screen or frames them oddly. In his blog post, Bordwell notes that Von Trier does not always choose the most outrageous framing generated by Automavision indicating that he realises that the unconventional framings have a particular effect on the viewer and should be mixed with more conventional compositions to create the intended viewing experience. If Automavision is truly random, as implied by Von Trier then there is no way to control the degree to which the resulting framings are unconventional. This means that Von Trier must keep refilming and regenerating framings with Automavision in order to get suitable shots.

The next generation Automavision system could improve its ability to generate unconventional framings by incorporating some of the intelligence of Virtual Cameras. If it is able to apply the classical framing conventions then it can knowingly violate them. This would also allow it to modify the conventionality of its framings by varying the relative influence of chance and the framing conventions. To do this it would need to begin processing the visual scene, which is no easy task and deciding where the significant objects should be positioned in the frame. An interesting reverse application of this technique would then be to apply Automavision 2.0 to a videogame. Anybody up for a game of Halo directed by Lars Von Trier?