# Focus on What Matters: Constraining Spatial-Temporal Attention via Action-Units for Noise-Resilient AQA

arXiv: https://arxiv.org/abs/2511.05611 · HTML: https://mlsanity.com/paper/2511.05611/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 21 of its 41 references that are in the map.
- First posted on arXiv 6 Nov 2025.

## Builds on (most cited first, 21 of 21)

- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset](https://mlsanity.com/paper/1705.07750/) (2017)
- [MS-TCN: Multi-Stage Temporal Convolutional Network for Action Segmentation](https://mlsanity.com/paper/1903.01945/) (2019)
- [ASFormer: Transformer for Action Segmentation](https://mlsanity.com/paper/2110.08568/) (2021)
- [FineDiving: A Fine-grained Dataset for Procedure-aware Action Quality Assessment](https://mlsanity.com/paper/2204.03646/) (2022)
- [What and How Well You Performed? A Multitask Learning Approach to Action Quality Assessment](https://mlsanity.com/paper/1904.04346/) (2019)
- [Learning To Score Olympic Events](https://mlsanity.com/paper/1611.05125/) (2016)
- [Alleviating Over-segmentation Errors by Detecting Action Boundaries](https://mlsanity.com/paper/2007.06866/) (2020)
- [Uncertainty-aware Score Distribution Learning for Action Quality Assessment](https://mlsanity.com/paper/2006.07665/) (2020)
- [Group-aware Contrastive Regression for Action Quality Assessment](https://mlsanity.com/paper/2108.07797/) (2021)
- [Learning to score the figure skating sports videos](https://mlsanity.com/paper/1802.02774/) (2018)
- [FineAction: A Fine-Grained Video Dataset for Temporal Action Localization](https://mlsanity.com/paper/2105.11107/) (2021)
- [Fine-grained Temporal Contrastive Learning for Weakly-supervised Temporal Action Localization](https://mlsanity.com/paper/2203.16800/) (2022)
- [FineParser: A Fine-grained Spatio-temporal Action Parser for Human-centric Action Quality Assessment](https://mlsanity.com/paper/2405.06887/) (2024)
- [Action Quality Assessment via Hierarchical Pose-guided Multi-stage Contrastive Regression](https://mlsanity.com/paper/2501.03674/) (2025)
- [Fine-grained Activity Recognition in Baseball Videos](https://mlsanity.com/paper/1804.03247/) (2018)
- [Auto-Encoding Score Distribution Regression for Action Quality Assessment](https://mlsanity.com/paper/2111.11029/) (2021)
- [CoFInAl: Enhancing Action Quality Assessment with Coarse-to-Fine Instruction Alignment](https://mlsanity.com/paper/2404.13999/) (2024)
- [Hierarchical NeuroSymbolic Approach for Comprehensive and Explainable Action Quality Assessment](https://mlsanity.com/paper/2403.13798/) (2024)
- [TimeMIL: Advancing Multivariate Time Series Classification via a Time-aware Multiple Instance Learning](https://mlsanity.com/paper/2405.03140/) (2024)
- [Multi-Stage Contrastive Regression for Action Quality Assessment](https://mlsanity.com/paper/2401.02841/) (2024)

## Abstract (arXiv)

The core challenge in Action Quality Assessment (AQA) lies in extracting fine-grained motion features from redundant and complex video backgrounds. Existing global feature learning methods are constrained by extremely low "signal-to-noise ratios", making it difficult to distinguish intrinsic actions from background clutter. To address this, we propose a Pose-Guided Intrinsic Motion Distillation Framework that explicitly enforces physical constraints to focus on motion subjects and decouple motion execution from environmental outcomes. First, we design an Action-Unit Parser that constructs dynamic regions of interest (ROIs) using human pose topology as prior knowledge. This functions as a spatial hard-attention filter that physically removes background noise at the input stage, forcing the model to learn appearance and geometric features only from pure body regions. Second, to resolve factor entanglement, we introduce a dual-stream decoupling mechanism: the Motion Parser focuses on capturing purified joint motion details, while the Condition Parser independently processes non-body-related environmental feedback (e.g., splash in diving) to create two orthogonal evaluation dimensions in feature space. Finally, adaptive weight modules integrate these decoupled features to generate final scores. Experimental results on large-scale datasets including FineDiving, FineDiving-HM, and MTL-AQA demonstrate that this method achieves state-of-the-art (SOTA) performance in both action segmentation and scoring accuracy, validating the effectiveness of "noise suppression focusing" and "motion disentanglement" strategies in fine-grained action evaluation.