ALGORITHM AS EXPERIMENTAL ART
The perspective proposed here regards artistic practice as an ongoing “elaboration on variable factors”.[4 : Focillon, 70.] To apply this paradigm to computational art means that the artist approaches every algorithm as a source of variable factors generative of multiple forms.
There are two ways of looking at algorithms, which can be termed instrumental and procedural.
The instrumental perspective is concerned with the algorithm’s outcome. This is the perspective of a “user” or “client” who needs a practical solution for the problem that the algorithm was designed to tackle: data compression, motion tracking, anti-aliasing, etc. The user only cares about whether the algorithm can solve the problem accurately or efficaciously. The details of how the problem is actually solved are from this standpoint not important. The algorithm is thus viewed as a means to an end, and the means is only important insofar as it accomplishes the end. From this performance-oriented or output-oriented standpoint, the algorithm remains a black box whose internal operation need not be known. Many digital artists use software libraries this instrumental way.
In contrast, the procedural perspective is concerned with the internal structure of the algorithm. This approach gives the standpoint of the scientist or programmer who wants to know the procedural details, possibly with the aim of writing her/his own implementation. A programmer often regards an algorithm as the expression of a way of thinking whose logic has to be understood. To understand the algorithm is not only to understand how it works but also to appreciate what makes it elegant and ingenious. Thus the programmer who knows the inner working of the technology also comes to admire it as a paradigm of intelligence and precision. The algorithmic artist adopts this second standpoint, but not in order to solve an engineering problem. Echoing Focillon’s paradigm of experimental art, the artist’s aim is to bring out the aesthetic possibilities of the algorithm as a formal generator. Flowpoints exemplifies this procedural mode of working.
The execution of LK proceeds by generating, testing, and refining hypotheses about the possible displacement of a template. The procedure proposes a possible motion vector, computes the error, and generates a refined motion vector. Thus the algorithm continually produces and discards hypothetical motion vectors. The artist who writes her/his implementation of the algorithm can use every vector produced in this iterative process.
The Flowpoints project consists of a series of line renderings produced by drawing the magnitude of every hypothesized motion vector, and to color code each line according to the depth of the current iteration. The earlier iterations generate lines drawn in warmer colors. The color becomes colder in the later iterations (Figure 4).
[ Figure 4 ]
There are several parameters that determine the graphical effect of each image. The major ones are the following:
- The size of each flowpoint is allowed to range from 4 by 4 to 50 by 50. Flowpoints are here assumed to be squares rather than rectangles.
- The distance between flowpoints can be changed, packing flowpoints so tightly that they overlap or situating them far from one another. Different images have been generated with different values of the distance parameter.
- For every flowpoint, the search continues looking for a match until a maximum number of iterations has been reached. Our experiments have allowed the iterations to range from 1 to 50.
- The fourth factor is the length of the vectors to be drawn. Some implementations of LK assume that motion vectors longer or shorter than a certain length are probably spurious and so discard them. The maximum and minimum length of the vectors to be drawn is subject to variation. In many of our experiments, the vectors have all been drawn, regardless of their length. The resulting images are thereforerichly textured.
The main procedure in this project is to select a movie clip and try different values of the above parameters in order to generate linear renderings that can be viewed online or displayed as digital prints.
The clips used here are all from the Andy Warhol movie Blow-Job. It consists of several long takes depicting a close-up of a young man. There is no camera movement or mobile framing of any sort (no tracking, dollying, panning, or tilting). In the context of experimental film history, Warhol’s cinematic method is important because it exemplifies a wider shift away from extensive usage of montage (rapid editing) towards very long takes with minimal editing. Because of its title (and information supplied in print by Warhol), we can assume that the man is receiving a blow-job, but the main focus of interest is in the duration of the event and especially the movements of the young man’s face. Warhol’s avoidance of rapid cutting directs attention in a sustained and concentrated way to the micro-rhythms of the human face. The fixed camera does not introduce any additional movement over and above that of the face. Even when cuts do occur, the camera remains in the same position across them. This method exemplifies film theorist and philosopher Béla Balázs’s emphasis on the power of the cinematic close-up to capture the micro-physiognomy of the human face.
The Flowpoints project proposes an alternative way of approaching the micro-movements of an image. The LK procedure, since it focuses on the movements of small windows of pixels, is particularly apt to search for very small changes in an image sequence. The following archive of materials shows some of the forms generated through experimentation with the Warhol movie. Echoing Focillon’s dictum, this archive is meant to record the score of experiments that constitutes this project.