12 upvotes
New Transformer Tracks Shifting Object Relations in Video
What is RTAformer: Relation-Transition-Aware Transformer for Video Scene Graph Generation?
RTAformer builds a scene graph for each video frame while explicitly tracking how object relationships appear, disappear, or change over time. It keeps object identities consistent across frames and highlights abrupt relationship shifts for clearer video understanding.
Who it is for
Anyone developing video analytics, search, or activity-recognition systems can use it to model dynamic interactions more accurately.