281201 Not-to-be-forgotten notes from discussions Jacco and I(Lynda) had. === Couple of suggestions for global constraints. --- 170901 Global constraints: Temp: audio single track or voice over and background track video: single track only === How "black" should our "black boxes" be in Cuypers? --- 190901 White box at current level, grey box when child, black box when grandchild. Bounding box too simple. Next least complex step, a sequence of bounding boxes? A left-of B => bounding box also in t Transitions (t) == alignement (x or y) === 210901 What is the problem with the labelling process? Are there underconstrained issues that should have been picked up before reaching this stage? Why would any labelling algorithm be better than another "random" one? We should look at the labelling process initially only as a means of looking more closely at the constraints that have been used. It's the "junk heap" at the end of all the explicitly encoded stuff. Have looked at a "white box" 3x3 slide show. What about a time-aligned NxN? Then, what about NxM - no implicit parallelism. Example of non-parallel is main slide show on user topic and smaller adverts in a corner. "Old" demo had problem with bounding boxes. The problem is at the boundary of 2 objects. Can you fit one black box (or media item) at the beginning of a white blob, and have the result be a white blob or a black box? Same goes for the end. Can we fit a black box into the end of a white blob? Does this mean that we need to try out white blobs as a construct in the code? -----