Planning and Generation
14 · Action Remaining Prediction
Action remaining prediction estimates how many seconds remain until the current basic surgical action ends, using observations available so far.
Input and output: The observed clip and current action information → the remaining duration in seconds.
Source data and labels: Surgical video with an action class, start time s and end time e for each action clip.
Construction method: Set an observation cutoff t within the action. Use footage from s to t as the observed clip. The remaining-time target is e − t; footage after t is not included in the input.
Figure 4. The input is the observed clip. The interval from t to e determines the remaining-time target.
Q: How many seconds remain before the action in the observed clip ends?
A: Remaining time = e − t seconds.