AJ Piergiovanni, Michael S. Ryoo
In this paper, we propose a convolutional layer inspired by optical flow algorithms to learn motion representations. Our representation flow layer is a fully-differentiable layer designed to capture the `flow' of any representation channel within a convolutional neural network for action recognition. Its parameters for iterative flow optimization are learned in an end-to-end fashion together with the other CNN model parameters, maximizing the action recognition performance. Furthermore, we newly introduce the concept of learning `flow of flow' representations by stacking multiple representation flow layers. We conducted extensive experimental evaluations, confirming its advantages over previous recognition models using traditional optical flows in both computational speed and performance. Code/models available here: https://piergiaj.github.io/rep-flow-site/
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Video | Kinetics-400 | Acc@1 | 77.9 | RepFlow-50 ([2+1]D CNN, FcF, Non-local block) |
| Activity Recognition | HMDB-51 | Average accuracy of 3 splits | 81.1 | RepFlow-50 ([2+1]D CNN, FcF, Non-local block) |
| Action Recognition | HMDB-51 | Average accuracy of 3 splits | 81.1 | RepFlow-50 ([2+1]D CNN, FcF, Non-local block) |