Long-Tail Temporal Action Segmentation with Group-wise Temporal Logit Adjustment
This addresses the challenge of recognizing infrequent actions in temporal segmentation for applications like video analysis, though it appears incremental as it builds on existing long-tail methods.
The paper tackles the problem of long-tailed action distribution in procedural activity videos, where existing methods fail to recognize tail actions, and proposes a G-TLA framework that significantly improves tail action segmentation without performance loss on head actions.
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.