Continual Action Quality Assessment via Adaptive Manifold-Aligned Graph Regularization

Name: Continual Action Quality Assessment via Adaptive Manifold-Aligned Graph Regularization
Uploaded: 2026-10-10T04:10:09.268214+00:00
Duration: 338 s
Description: Continual AQA addresses challenges in action quality assessment where real-world score distributions change over time. The proposed MAGR++ framework uses layer-adaptive fine-tuning and a Manifold Projector, combined with graph regularization, to balance capturing fine motion cues with maintaining stability across evolving data distributions. It achieves state-of-the-art performance.

Continual AQA addresses challenges in action quality assessment where real-world score distributions change over time. The proposed MAGR++ framework uses layer-adaptive fine-tuning and a Manifold Projector, combined with graph regularization, to balance capturing fine motion cues with maintaining stability across evolving data distributions. It achieves state-of-the-art performance.

Open the full episode