Exploring Adaptive Masked Reconstruction for Self-Supervised Skeleton-Based Action Recognition
Researchers have developed a new framework called Adaptive Masked Reconstruction (AMR) to improve self-supervised skeleton-based action recognition. AMR speeds up the pre-training process by decoupling the decoder from the encoder, allowing for more flexible prediction of larger spatiotemporal patches. It also incorporates an adaptive guidance module that directs the model's focus to the most informative motion patterns, enhancing recognition accuracy. Experiments show AMR significantly reduces training time and outperforms existing state-of-the-art methods on multiple benchmark datasets. AI
IMPACT Accelerates training and improves accuracy for skeleton-based action recognition tasks.