In order to successfully generate graphs that represent interactions between actors, we require data that is not necessarily very detailed for each time interval, but rather data that is available for very short intervals. Our goal in devising Interaction Graphs is to eventually be able to make a judgment on how much interaction there is between two given actors. Had we only 14 scenes and 2 interaction summaries per scene, we may be able to judge about what actor interacts with what other actor; however, our graph will lack information on the degree of interaction. In terms of amount of information per time interval, this very first version of VideoFace makes use of 2 types of data - 1. actor names in a given interval, and 2. location of a given interval. These two types suffice in generating an ActorGraph and an ActorSceneLocationGraph.

Information about locations should be sparse, that is there should be no more than 10 distinct locations used in the ShotSpreadSheet. The reason for this becomes apparent when examining the ActorSceneLocationGraph. The graph has been divided into rows that correspond to locations, and columns that correspond to scenes. The larger the number of locations (= rows), the smaller becomes the height of each row, and consequently the height of each detailed graph that is located in table cells. Similarly, the number of scenes should not exceed a certain number as well. Since screen width is typically on the order of 1/3 larger than screen height, the number of scenes can be 1/3 more than the number of locations, i.e. on the order of a maximum of 15. These values were measured for a screen size of 1280*1024, and they may vary for higher resolutions.

So far, only one particular layout of a ShotSpreadSheet has been considered. It is not too difficult to extend the implemented version to a more general scheme, as discussed in the section titled "Speculation on Extensions". A ShotSpreadSheet at this point is expected to contain 11 columns, where column 7 contains actor information and column 11 contains location information.

Figure 4: Isolation Mode
Click to enlarge

Moreover, every scene is expected to begin with the title of the scene in the first column while all other columns are not to contain information. While location information is rather flexible, actor information is more restrictive. An actor name must be no shorter than 2 characters; the first character must necessarily be a letter, and the last character must be a digit. Whether the actor's name is 'abcdef12345' or 'm1' makes no difference.