We propose a unified and generalizable framework for object state recognition in video data.

TopER is a low-dimensional, interpretable graph embedding method based on topological evolution rates, enabling intuitive visualization and strong performance on clustering and classification tasks.