Pistachio: Towards Synthetic, Balanced, and Long-Form Video Anomaly Benchmarks

2025-11-22

AI Summary

Existing VAD/VAU benchmarks suffer from limited scene diversity, imbalanced anomaly class distribution, and insufficient temporal complexity; moreover, VAU demands deep semantic understanding and causal reasoning, rendering manual annotation costly and unscalable. To address these limitations, we introduce Pistachio—the first fully generative video anomaly benchmark—built upon state-of-the-art video diffusion models. It employs scene-conditioned anomaly injection, multi-stage narrative planning, and long-sequence temporal consistency optimization to synthesize coherent, controllable 41-second abnormal videos. Pistachio enables diverse, balanced, and long-horizon (multi-event) anomaly modeling, effectively mitigating real-data biases and annotation bottlenecks. Experiments demonstrate that Pistachio surpasses existing benchmarks in scale, diversity, and temporal complexity, exposing critical weaknesses of mainstream methods in dynamic, long-duration anomaly understanding. It establishes a new standard and identifies key challenges for future VAU research.

Technology Category

• Computer Vision: Video Understanding & Activity Analysis
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Automatically detecting abnormal events in videos is crucial for modern autonomous systems, yet existing Video Anomaly Detection (VAD) benchmarks lack the scene diversity, balanced anomaly coverage, and temporal complexity needed to reliably assess real-world performance. Meanwhile, the community is increasingly moving toward Video Anomaly Understanding (VAU), which requires deeper semantic and causal reasoning but remains difficult to benchmark due to the heavy manual annotation effort it demands. In this paper, we introduce Pistachio, a new VAD/VAU benchmark constructed entirely through a controlled, generation-based pipeline. By leveraging recent advances in video generation models, Pistachio provides precise control over scenes, anomaly types, and temporal narratives, effectively eliminating the biases and limitations of Internet-collected datasets. Our pipeline integrates scene-conditioned anomaly assignment, multi-step storyline generation, and a temporally consistent long-form synthesis strategy that produces coherent 41-second videos with minimal human intervention. Extensive experiments demonstrate the scale, diversity, and complexity of Pistachio, revealing new challenges for existing methods and motivating future research on dynamic and multi-event anomaly understanding.

Problem

• Existing VAD benchmarks lack diversity and balanced anomaly coverage
• Video Anomaly Understanding requires deeper reasoning but lacks benchmarks
• Manual annotation for VAU benchmarks demands heavy human effort

Innovation

• Generation-based pipeline for video anomaly benchmarks
• Scene-conditioned anomaly assignment and storyline generation
• Temporally consistent long-form synthesis strategy