提升骨骼动作识别的长期与短期时序建模，同时增强数据多样性。
LSTC-MDA: A Unified Framework for Long-Short Term Temporal Convolution and Mixed Data Augmentation in Skeleton-Based Action Recognition
- 设计并行长短时分支的LSTC模块，自适应融合保留关键长程特征。
- 在NTU60/120和NW-UCLA上达到94.1%~97.5%的最高准确率。
- 适合关注骨骼动作识别中时序建模与数据增强的研究者。
基于骨骼的动作识别长期面临标注样本稀缺与短长期时序依赖建模困难的问题。为此，本文提出统一框架LSTC-MDA，同时优化时序建模与数据多样性。引入新型长-短时序卷积（LSTC）模块，包含并行的短程与长程分支，通过学习的相似性权重自适应对齐与融合，保留传统步长为2的时序卷积所丢失的关键长程信息。此外，将关节混合数据增强（JMDA）扩展至输入层的加性混叠（Additive Mixup），在同视角内进行混叠以避免分布偏移。消融实验验证各组件有效性。LSTC-MDA在NTU60（X-Sub: 94.1%，X-View: 97.5%）、NTU120（X-Sub: 90.4%，X-Set: 92.0%）和NW-UCLA（97.2%）上取得当前最优性能。代码已开源：https://github.com/xiaobaoxia/LSTC-MDA。
原文摘要 · Abstract (English)
Skeleton-based action recognition faces two longstanding challenges: the scarcity of labeled training samples and difficulty modeling short- and long-range temporal dependencies. To address these issues, we propose a unified framework, LSTC-MDA, which simultaneously improves temporal modeling and data diversity. We introduce a novel Long-Short Term Temporal Convolution (LSTC) module with parallel short- and long-term branches, these two feature branches are then aligned and fused adaptively using learned similarity weights to preserve critical long-range cues lost by conventional stride-2 temporal convolutions. We also extend Joint Mixing Data Augmentation (JMDA) with an Additive Mixup at the input level, diversifying training samples and restricting mixup operations to the same camera view to avoid distribution shifts. Ablation studies confirm each component contributes. LSTC-MDA achieves state-of-the-art results: 94.1% and 97.5% on NTU 60 (X-Sub and X-View), 90.4% and 92.0% on NTU 120 (X-Sub and X-Set),97.2% on NW-UCLA. Code: https://github.com/xiaobaoxia/LSTC-MDA.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。