Research map / Video / Video segmentation and tracking
Learning Event-Relevant Factors for Video Anomaly Detection
Che Sun, Chenrui Shi, Yunde Jia, Yuwei Wu
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add event-relevant factor learning and counterfactual finetuning to HF2-VAD | HF2-VAD · ShanghaiTech, Setting-A | AUROC: 76.2 -> 78.6 | Improved |
| Add event-relevant factor learning and counterfactual finetuning to HF2-VAD | HF2-VAD · CUHK Avenue, Setting-A | AUROC: 91.1 -> 91.5 | Improved |
| Add event-relevant factor learning and counterfactual finetuning to HF2-VAD | HF2-VAD · UCSD Ped2, Setting-A | AUROC: 99.3 -> 99.4 | Improved |
| Add event-relevant factor learning and counterfactual finetuning to HF2-VAD | HF2-VAD · ShanghaiTech, Setting-B | AUROC: 73.8 -> 76.4 | Improved |
| Add event-relevant factor learning and counterfactual finetuning to HF2-VAD | HF2-VAD · CUHK Avenue, Setting-B | AUROC: 82.2 -> 87.8 | Improved |
| Add event-relevant factor learning and counterfactual finetuning to HF2-VAD | HF2-VAD · UCSD Ped2, Setting-B | AUROC: 98.4 -> 99.2 | Improved |
| Use standard VAE instead of memory-augmented VAE with two representations (ablation) | Event-relevant factor learning with HF2-VAD · ShanghaiTech, Setting-B | AUROC: 74.2 -> 76.4 | Improved |
| Use memory-augmented VAE with one representation instead of two (ablation) | Event-relevant factor learning with HF2-VAD · ShanghaiTech, Setting-B | AUROC: 72.6 -> 76.4 | Improved |
| Use standard VAE with two representations instead of memory-augmented VAE (ablation) | Event-relevant factor learning with HF2-VAD · ShanghaiTech, Setting-B | AUROC: 75.0 -> 76.4 | Improved |
| Add the causal information-flow loss (ablation) | Memory-augmented VAE with two representations and counterfactual finetuning · ShanghaiTech, Setting-B | AUROC: 74.1 -> 76.4 | Improved |
| Add the causal information-flow loss (ablation) | Memory-augmented VAE with two representations and counterfactual finetuning · CUHK Avenue, Setting-B | AUROC: 85.8 -> 87.8 | Improved |
| Use additive interventions instead of dropout-based interventions for counterfactual generation (ablation) | Memory-augmented VAE with two representations and causal loss · ShanghaiTech, Setting-B | AUROC: 75.2 -> 76.4 | Improved |
Closest papers by meaning
- A Causal Inference Look at Unsupervised Video Anomaly Detection (AAAI 2022 · similarity 0.92)
- A Discriminative Framework for Anomaly Detection in Large Videos (ECCV 2016 · similarity 0.89)
- Towards Video Anomaly Detection from Event Streams: A Baseline and Benchmark Datasets (ECCV 2026 · similarity 0.88)
- Future Frame Prediction for Anomaly Detection – A New Baseline (CVPR 2018 · similarity 0.88)
- VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree (NeurIPS 2025 · similarity 0.87)
- Video Anomaly Detection with Motion and Appearance Guided Patch Diffusion Model (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
