Transformer improves detection of changing relationships in video scenes
RTAformer: Relation-Transition-Aware Transformer for Video Scene Graph Generation
Computer Vision and Pattern Recognition
Summary
Videos show objects and how they relate to each other over time, but sometimes relationships change quickly and models can get stuck thinking the old relationship still exists. The authors created a transformer-based method called RTAformer that treats objects staying the same and relationships changing separately. This helps the model update relationship predictions when needed without mixing up object identities. Their tests show this approach better captures switches in relationships between objects in videos.
What this means in practice
- •For video surveillance teams: Detect dynamic interactions between objects more accurately in video feeds for events requiring rapid relationship updates.
- •For robotics developers: Improve a robot’s understanding of object relationships changes in real time to better navigate and interact in dynamic environments.
Authors
Jaeyeon Heo, Geuntaek Lim, Yukyung Choi
Abstract
Video scene graph generation (VidSGG) aims to capture objects and their evolving relationships in videos. This requires accounting for both the persistence of object identities and abrupt changes in relationships. Existing methods aggregate information across frames to exploit temporal context, but an emphasis on consistency can obscure these changes. As a result, a model may retain its previous prediction even after the underlying relation has changed. We refer to this failure mode as relational semantic inertia. To address this issue, we propose Relation-Transition-Aware Transformer (RTAformer), a one-stage framework that separately models entity persistence and relation changes. It injects calibrated differences between adjacent-frame relation queries to guide predicate updates, while cross-frame instance alignment during training helps reduce cross-frame representational drift. Experiments on Action Genome show improvements in recall and mean recall over existing methods, with ablation studies supporting the contributions of the proposed components and comparisons using matched backbones showing larger gains on relation-transition frames than on non-transition frames.