Recognize Complex Events from Static Images by Fusing Deep Channels Supplementary Materials

@inproceedings{Xiong2015RecognizeCE,
  title={Recognize Complex Events from Static Images by Fusing Deep Channels Supplementary Materials},
  author={Yuanjun Xiong and Kai Zhu and Dahua Lin and Xiaoou Tang},
  year={2015}
}
RCNNBank The RCNNBank is implemented based on ObjectBank [3], except that the detector responses are replaced by the activation features obtained using CNN. For each image, we first obtain 500 bounding boxes with highest proposal scores and apply the CNN to derive action features. This results in 500 activation feature vectors of 4096 dimensions. Then we… CONTINUE READING