Official PyTorch implementation of "Less is More: Decoder-Free Masked Modeling for Efficient Skeleton Representation Learning".

SLiM (Skeleton Less is More) is a unified representation learning framework that bridges the gap between Masked Auto-Encoders (MAE) and Contrastive Learning (CL).

• 🚀 Extreme Efficiency: A completely decoder-free architecture that reduces inference computational costs by 7.89×.
• 🧠 Robust Representation: Introduces Semantic Tube Masking (STM) and Skeleton-Aware Augmentations (SAA) to capture anatomically consistent motion dynamics without shortcut learning.
• 🏆 State-of-the-Art: Achieves peak performance across NTU-60, NTU-120, and PKU-MMD II benchmarks.

• Mar 11, 2026: This repository is created

@article{do2026less,
  title={Less is More: Decoder-Free Masked Modeling for Efficient Skeleton Representation Learning},
  author={Do, Jeonghyeok and Chen, Yun and Youk, Geunhyuk and Kim, Munchurl},
  journal={arXiv preprint arXiv:2603.10648},
  year={2026}
}

Please visit our project page for more experimental results.

This repository is built upon SkateFormer and TDSM.