ol8Dsa

bocsitboals

Is there any measure of how well the object level grouping algorithm works (precision-recall, etc.)? I would particularly be interested to see if it comes up with any false positives, and what those might look like. They seem to be able to detect the van and Bill Murray (Figs 10+11) very well. It seems like they might be able to find Bill if we threw in some frames from Ghostbusters as well. Also did they mention how long any of this takes?

Very interesting. This seems to work quite well in the examples presented. They mention that the "words" trained from Run Lola Run where also used satisfactorily on Groundhog Day, but I'm not sure how this can be. This seems to be more of a single coherent video method at the moment. My main questions have to do with the conversion to text retrival as I'm not familiar with this at all. The conversion from object frequencies to a suitable search vector is lost on me.