post

发表于 分类于 阅读次数：
Enhancing Action Quality Assessment with Coarse-to-Fine Instruction Alignment

IJCAI 2024

论文

Abstract

the inherent struggle of these backbones to capture the subtle cues essential for AQA,gains of 5.49% and 3.55% on Rhythmic Gymnastics and Fis-V.

Introduction

involves leveraging the backbone pre-trained on large-scale action recognition datasets to adapt the score regression requirements of small-scale AQA datasets.

two critical issues: domain shift and a severe risk of overfitting.

characterizing AQA as a coarse-to-fine classification task
using an MLP classifier to predict the coarse-grained grades, a fine-grained classification task to discern variations within each grade.

ST-GCN, 预训练的模型存在差异。

present an innovative solution—a novel pre-training alignment for AQA by aligning it with the pre-trained task.

Neural Collapse(fall): study AQA from the neural collapse perspective, has interpretability(understand, explain).

Methodology

代码