Abstract
The Fisher vector (FV) representation is a highdimensional extension of the popular bag-of-word representation. Transformation of the FV by power and ℓ2 normalizations has shown to signifificantly improve its performance, and led to state-of-the-art results for a range of image and video classifification and retrieval tasks. These normalizations, however, render the representation non-additive over local descriptors. Combined with its high dimensionality, this makes the FV computationally expensive for the purpose of localization tasks. In this paper we present approximations to both these normalizations, which yield signififi- cant improvements in the memory and computational costs of the FV when used for localization. Second, we show how these approximations can be used to defifine upper-bounds on the score function that can be effificiently evaluated, which enables the use of branch-and-bound search as an alternative to exhaustive sliding window search. We present experimental evaluation results on classifification and temporal localization of actions in videos. These show that the our approximations lead to a speedup of at least one order of magnitude, while maintaining state-of-the-art action recognition and localization performance.