Learn More
Convolutional neural networks (CNNs) are inherently limited to model geometric transformations due to the fixed geometric structures in its building modules. In this work, we introduce two new modules to enhance the transformation modeling capacity of CNNs, namely, deformable con-volution and deformable RoI pooling. Both are based on the idea of augmenting(More)
Deep convolutional neutral networks have achieved great success on image recognition tasks. Yet, it is non-trivial to transfer the state-of-the-art image recognition networks to videos as per-frame evaluation is too slow and un-affordable. We present deep feature flow, a fast and accurate framework for video recognition. It runs the expensive convolutional(More)
  • 1