an entry from

Piotr's R&D blog

At Peggy's suggestion, I exhibited the Video Bench at the UIST reception hosted by NewMIC. It went over well, especially considering that the audience consisted of experts in user interface technologies and that there were 38 other competing exhibits on the floor. I don't have any pictures, but here's the list of ideas from the visitors:

• IBM has a toolkit that does scene boundary detection. I checked alphaWorks, and apparently CueVideo has been retired, but the MPEG-7 Annotation Tool has similar capabilities. We'll have to integrate it into Video Bench.
• Though the DT surfaces are getting larger, they're still not big enough. It would be nice to be able to put together many tables into a larger mosaic, and use it as one virtual surface, kind of like extended desktop in multi-monitor mode on Windows XP. This would be easier to do if all tabletops were connected to one machine, but it would be interesting to apply Brian's distributed variation to this problem.
• There were the usual complaints about the lack of transitions and special effects in Video Bench. One person suggested that fades could be done by cutting a cel at an angle; I guess the fade would either be of fixed duration, or perhaps depend on the angle of the cut. This would take care of 90% of useful transition effects, and integrates well with existing gestures.
• Somebody else noted that there is little context when scrubbing through video, so for example it's difficult to pick a specific frame on which to cut. I think we might be able to fix this by enabling a special spread mode where you spread out one cel into three: the central cel contains only one frame, and the ones on either side show the immediately adjacent frames. When scrubbing through a one-frame cel, special semantics apply that shift frames from adjacent cels through the central one.
• The usual problem of mistaken splices (usually into the non-undoable trashcan) cropped up. The problem is, of course, that the user's focus is where he grabbed the strip, which is often nowhere near the two ends whose location causes actions to be taken. Highlighting to indicate that an action will occur is not sufficient. The movement gesture must be revisited. Perhaps if we get rid of jogging (another dismal failure), we can use single-finger dragging instead? Then you must drag at a divider to make it active, otherwise the strip won't interact with anything.

I also got to see a preview of some other DT work, but sadly didn't have time to roam the reception floor to see all the other cool exhibits. Shame.