Cewu Lu, Ranjay Krishna, Michael Bernstein, Li Fei-Fei
Visual relationships capture a wide variety of interactions between pairs of objects in images (e.g. "man riding bicycle" and "man pushing bicycle"). Consequently, the set of possible relationships is extremely large and it is difficult to obtain sufficient training examples for all possible relationships. Because of this limitation, previous work on visual relationship detection has concentrated on predicting only a handful of relationships. Though most relationships are infrequent, their objects (e.g. "man" and "bicycle") and predicates (e.g. "riding" and "pushing") independently occur more frequently. We propose a model that uses this insight to train visual models for objects and predicates individually and later combines them together to predict multiple relationships per image. We improve on prior work by leveraging language priors from semantic word embeddings to finetune the likelihood of a predicted relationship. Our model can scale to predict thousands of types of relationships from a few examples. Additionally, we localize the objects in the predicted relationships as bounding boxes in the image. We further demonstrate that understanding relationships can improve content based image retrieval.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Scene Parsing | VRD Relationship Detection | R@100 | 14.7 | Lu et. al [[Lu et al.2016]] |
| Scene Parsing | VRD Relationship Detection | R@50 | 13.86 | Lu et. al [[Lu et al.2016]] |
| Scene Parsing | VRD Predicate Detection | R@100 | 47.87 | Lu et. al [[Lu et al.2016]] |
| Scene Parsing | VRD Predicate Detection | R@50 | 47.87 | Lu et. al [[Lu et al.2016]] |
| Scene Parsing | VRD Phrase Detection | R@100 | 17.03 | Lu et. al [[Lu et al.2016]] |
| Scene Parsing | VRD Phrase Detection | R@50 | 16.17 | Lu et. al [[Lu et al.2016]] |
| Scene Parsing | VRD | Recall@50 | 18.16 | VRD |
| Visual Relationship Detection | VRD Relationship Detection | R@100 | 14.7 | Lu et. al [[Lu et al.2016]] |
| Visual Relationship Detection | VRD Relationship Detection | R@50 | 13.86 | Lu et. al [[Lu et al.2016]] |
| Visual Relationship Detection | VRD Predicate Detection | R@100 | 47.87 | Lu et. al [[Lu et al.2016]] |
| Visual Relationship Detection | VRD Predicate Detection | R@50 | 47.87 | Lu et. al [[Lu et al.2016]] |
| Visual Relationship Detection | VRD Phrase Detection | R@100 | 17.03 | Lu et. al [[Lu et al.2016]] |
| Visual Relationship Detection | VRD Phrase Detection | R@50 | 16.17 | Lu et. al [[Lu et al.2016]] |
| Scene Understanding | VRD Relationship Detection | R@100 | 14.7 | Lu et. al [[Lu et al.2016]] |
| Scene Understanding | VRD Relationship Detection | R@50 | 13.86 | Lu et. al [[Lu et al.2016]] |
| Scene Understanding | VRD Predicate Detection | R@100 | 47.87 | Lu et. al [[Lu et al.2016]] |
| Scene Understanding | VRD Predicate Detection | R@50 | 47.87 | Lu et. al [[Lu et al.2016]] |
| Scene Understanding | VRD Phrase Detection | R@100 | 17.03 | Lu et. al [[Lu et al.2016]] |
| Scene Understanding | VRD Phrase Detection | R@50 | 16.17 | Lu et. al [[Lu et al.2016]] |
| 2D Semantic Segmentation | VRD Relationship Detection | R@100 | 14.7 | Lu et. al [[Lu et al.2016]] |
| 2D Semantic Segmentation | VRD Relationship Detection | R@50 | 13.86 | Lu et. al [[Lu et al.2016]] |
| 2D Semantic Segmentation | VRD Predicate Detection | R@100 | 47.87 | Lu et. al [[Lu et al.2016]] |
| 2D Semantic Segmentation | VRD Predicate Detection | R@50 | 47.87 | Lu et. al [[Lu et al.2016]] |
| 2D Semantic Segmentation | VRD Phrase Detection | R@100 | 17.03 | Lu et. al [[Lu et al.2016]] |
| 2D Semantic Segmentation | VRD Phrase Detection | R@50 | 16.17 | Lu et. al [[Lu et al.2016]] |
| 2D Semantic Segmentation | VRD | Recall@50 | 18.16 | VRD |
| Scene Graph Generation | VRD | Recall@50 | 18.16 | VRD |