Large displacement optical flow & flow-based action recognition

In the first part of our talk, we present an approach for large displacement optical flow. Optical flow computation is a key component in many computer vision systems designed for tasks such as action
detection or activity  recognition. Inspired by the large displacement optical flow of Brox and  Malik, our approach  DeepFlow  combines a novel matching algorithm with a variational approach . Our matching algorithm builds upon a multi-stage architecture interleaving convolutions and max-pooling.  DeepFlow efficiently handles large displacements  occurring in realistic videos, and shows competitive performance on optical flow benchmarks.

In the second part of our talk, we present a state-of-the-art approach  for action recognition based  on motion stabilized trajectory  descriptors and a Fisher vector representation.  We briefly review the recent trajectory-based video features and, then, introduce their motion stabilized version, combining human detection and dominant motion estimation. Fisher vectors summarize the information of a video efficiently. Results on several of the recent action datasets as well as the TrecVid MED dataset show that our approach outperforms the state-of-the-art