RelateAnything: Predicting Relationships in Images
RelateAnything is a real-time model for predicting relationships between objects in images with an open predicate vocabulary.
How the system works
The model takes an image and regions of interest from any source, returning estimated relationships based on predicates defined during inference as strings.
Key features of RelateAnything
- Does not require object labels as input data
- Allows using various region sources without retraining
- Works with an open predicate vocabulary











