Hierarchical Action Classification with Network Pruning

TL;DR

This paper introduces a hierarchical action classification approach that leverages network pruning and skeleton preprocessing to improve robustness and performance across multiple datasets, establishing new benchmarks especially on NTU 120.

Contribution

It presents a novel combination of hierarchical classification, network pruning, and skeleton preprocessing to enhance human action recognition performance.

Findings

Achieves comparable or superior results on four datasets.

Sets a new baseline for NTU 120 dataset.

Provides extensive analysis and ablation studies.

Abstract

Research on human action classification has made significant progresses in the past few years. Most deep learning methods focus on improving performance by adding more network components. We propose, however, to better utilize auxiliary mechanisms, including hierarchical classification, network pruning, and skeleton-based preprocessing, to boost the model robustness and performance. We test the effectiveness of our method on four commonly used testing datasets: NTU RGB+D 60, NTU RGB+D 120, Northwestern-UCLA Multiview Action 3D, and UTD Multimodal Human Action Dataset. Our experiments show that our method can achieve either comparable or better performance on all four datasets. In particular, our method sets up a new baseline for NTU 120, the largest dataset among the four. We also analyze our method with extensive comparisons and ablation studies.