Long-Tail Temporal Action Segmentation with Group-wise Temporal Logit Adjustment

TL;DR

This paper introduces G-TLA, a novel framework for long-tail temporal action segmentation that effectively improves tail action recognition by leveraging activity information and action ordering, without sacrificing head action performance.

Contribution

The paper proposes a group-wise temporal logit adjustment framework that addresses long-tail challenges in temporal action segmentation by integrating activity context and action sequence information.

Findings

Significant improvement in tail action segmentation accuracy.

No performance loss on head actions.

Effective handling of long-tailed action distributions.

Abstract

Procedural activity videos often exhibit a long-tailed action distribution due to varying action frequencies and durations. However, state-of-the-art temporal action segmentation methods overlook the long tail and fail to recognize tail actions. Existing long-tail methods make class-independent assumptions and struggle to identify tail classes when applied to temporal segmentation frameworks. This work proposes a novel group-wise temporal logit adjustment~(G-TLA) framework that combines a group-wise softmax formulation while leveraging activity information and action ordering for logit adjustment. The proposed framework significantly improves in segmenting tail actions without any performance loss on head actions.