Work

Unified skeleton representation learning for autonomous driving

Year: 2026

Type: article

Abstract: • MotionMoCo learns generic skeleton representations that are adaptable to different downstream tasks. • MotionMoCo learns unified skeleton representations by decoupling spatial and temporal features. • A momentum contrastive framework ensures robust and transferable spatiotemporal embeddings. • Motion-Aware Frame Masking selectively removes fast-moving frames to enhance feature learning. • Achieves competitive results on action recognition, 2D-to-3D lifting, and human mesh recovery. • Enables multi-task generalization without requiring task-specific annotations or architectures.

Source: Transportation Research Part C Emerging Technologies

DOI: 10.1016/j.trc.2026.105753

Authors: Mohamed Abdelfattah, Alexandre Alahi

Institution: École Polytechnique Fédérale de Lausanne

Language: English

Topic: Human Pose and Action Recognition

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Estimated APC paid: $4,040

Funder: Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung