Pose-Aware Attention Network for Flexible Motion Retargeting by Body Part

Area of Science:

• Computer Graphics
• Computer Vision
• Animation

Background:

• Traditional motion retargeting methods impose strict constraints on skeleton topology and joint count.
• These limitations hinder retargeting across diverse character structures.

Purpose of the Study:

• To develop a novel and flexible motion retargeting framework.
• To overcome the limitations of existing methods by treating body parts as fundamental units.

Main Methods:

• Proposes a body-part-centric motion retargeting strategy.
• Introduces a pose-aware attention network (PAN) for enhanced spatial modeling in motion encoding.
• PAN dynamically predicts joint weights and constructs shared latent spaces for body parts.

Main Results:

• Achieves superior qualitative and quantitative motion retargeting results compared to state-of-the-art methods.
• Demonstrates effective retargeting between dissimilar skeletons, including bipedal and quadrupedal types.
• The body part strategy and PAN contribute to robust performance across challenging scenarios.

Conclusions:

• The proposed framework offers a flexible and effective solution for motion retargeting.
• Body part-based retargeting and the pose-aware attention network are key innovations.
• The method significantly advances the capabilities of motion retargeting in computer graphics and vision.