Figure 1.

The architecture of 2S-RLSTM. Given a frame sequence and a label sequence, 2S-RLSTM can predict a series of actions and their durations in an iterative way.

The architecture of 2S-RLSTM. Given a frame sequence and a label sequence, 2S-RLSTM can predict a series of actions and their durations in an iterative way.