How do off-the-shelf scene detection or video segmentation algorithms work, and what are their typical failure modes on diverse video content?
