Description

Temporal Action Segmentation (TAS) deals with untrimmed procedural videos, where temporal relations between actions are prominent and may span over long temporal duration. The temporal context plays a significant role in this task. While in an offline setting, the full sequence is visible and consequently the temporal context can be captured relatively easily, in an online setting, it become a real challenge. This project aims at developing a new online framework for the underexplored task of online TAS.