BUSSARD: Normalizing Flows for Bijective Universal Scene-Specific Anomalous Relationship Detection

2026-03-17

AI Summary

This work addresses the problem of detecting anomalous semantic relationships between objects in scene graphs by proposing the first scene-specific anomaly detection framework based on bijective normalizing flows. The method integrates multimodal embeddings with semantic encodings from language models to map object–relation–object triplets onto a simple base distribution, enabling anomaly identification through likelihood estimation. Evaluated on the SARD dataset, the approach achieves an absolute improvement of approximately 10% in AUROC over the current state-of-the-art, while offering a fivefold increase in inference speed. Furthermore, it demonstrates enhanced robustness and generalization under semantic variations, exhibiting a 17.5% reduction in performance fluctuation when subjected to synonym perturbations.

Technology Category

• Computer Vision: Scene Analysis & Understanding
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding

Abstract

We propose Bijective Universal Scene-Specific Anomalous Relationship Detection (BUSSARD), a normalizing flow-based model for detecting anomalous relations in scene graphs, generated from images. Our work follows a multimodal approach, embedding object and relationship tokens from scene graphs with a language model to leverage semantic knowledge from the real world. A normalizing flow model is used to learn bijective transformations that map object-relation-object triplets from scene graphs to a simple base distribution (typically Gaussian), allowing anomaly detection through likelihood estimation. We evaluate our approach on the SARD dataset containing office and dining room scenes. Our method achieves around 10% better AUROC results compared to the current state-of-the-art model, while simultaneously being five times faster. Through ablation studies, we demonstrate superior robustness and universality, particularly regarding the use of synonyms, with our model maintaining stable performance while the baseline shows 17.5% deviation. This work demonstrates the strong potential of learning-based methods for relationship anomaly detection in scene graphs. Our code is available at https://github.com/mschween/BUSSARD .

Problem

• anomalous relationship detection
• scene graphs
• normalizing flows
• multimodal embedding
• semantic anomaly

Innovation

• Normalizing Flows
• Scene Graph Anomaly Detection
• Bijective Transformation
• Multimodal Embedding
• Likelihood-based Anomaly Detection