Neural Motifs: Scene Graph Parsing with Global Context
This work addresses scene understanding for computer vision by improving structured graph representations, with incremental gains over existing methods.
The paper tackled scene graph parsing by analyzing motifs (recurring substructures) in the Visual Genome dataset, finding that object labels predict relations and over 50% of graphs contain multi-relation motifs, leading to a baseline that improved state-of-the-art by 3.6% and a new architecture (Stacked Motif Networks) that further improved by 7.1%.
We investigate the problem of producing structured graph representations of visual scenes. Our work analyzes the role of motifs: regularly appearing substructures in scene graphs. We present new quantitative insights on such repeated structures in the Visual Genome dataset. Our analysis shows that object labels are highly predictive of relation labels but not vice-versa. We also find that there are recurring patterns even in larger subgraphs: more than 50% of graphs contain motifs involving at least two relations. Our analysis motivates a new baseline: given object detections, predict the most frequent relation between object pairs with the given labels, as seen in the training set. This baseline improves on the previous state-of-the-art by an average of 3.6% relative improvement across evaluation settings. We then introduce Stacked Motif Networks, a new architecture designed to capture higher order motifs in scene graphs that further improves over our strong baseline by an average 7.1% relative gain. Our code is available at github.com/rowanz/neural-motifs.