Extended KAFR: a kinematic-adaptive paradigm for the efficient analysis of surgical video

Each video is recorded at 25 frames per second (FPS) and manually annotated at 1 fps for surgical phase segmentation.

Cholec80 includes a wide range of intra-procedural variability, making it a robust benchmark for evaluating surgical workflow analysis methods, particularly in temporal modeling and surgical phase segmentation (Fig.

These methods represent recent advancements in surgical phase segmentation and have demonstrated strong performance on the Cholec80 benchmark dataset.

For comparison under strict evaluation (without 10-second relaxation), we used a baseline KAFR model with 1.5% frame retention.

This study presents KAFR, a framework for surgical phase segmentation that selects frames based on kinematic saliency rather than fixed temporal intervals.