An Iterative Classification Network for Semantic Action Recognition
Lijuan Marissa Zhou, Junfu Chen, Xiaojie Qian · 2021
Action recognition has been employed in many fields such as autonomous vehicles, surveillance, etc. A novel method for semantic action recognition through an iterative classification network is proposed in this paper. This paper assumes that action labels with noises are also beneficial for classification, the network iteratively applies the inferred classification results of previous iteration. The noisy action labels could be also corrected through multiple iterations. Specifically, the proposed iterative classification network improves the Convolutional Bi-directional Long Short-Term Memory (Bi-LSTM) network by adding iterative mechanism, which incorporates the classification probabilities in previous one iteration into the feature representation in the next one iteration. Experiments conducted on MSRC-12 and WorkoutUOW-18 datasets verify the effectiveness of the proposed method.