Sparse flexible models of local features. Gustavo Carneiro and David Lowe. Proc. ECCV 2006.
Includes a nice overview of different structures (e.g., trees, star graphs, etc.) that can be used to represent relations between parts.
Subscribe to:
Post Comments (Atom)
I'm curious about the magic parameter k. They have one plot of changes in accuracy for varying k, but they show it for faces which of course will see better results with more connections. If the idea behind using semilocal rigidity is to handle objects that are not rigid, I'd like to see performance metrics for varying levels of k for something like the snakes, or human bodies, which display more interesting properties. Any insight into this?
ReplyDeleteAlso, it's kinda weird that there's no tradeoff demonstrated (except computation time) for increasing k and the number of features in the model. that is, the EER plots are monotonic...where is the overfitting for increasing these values?
ReplyDelete