Research map / Video / Video segmentation and tracking
Mixture of Experts Guided by Gaussian Splatters Matters: A new Approach to Weakly-Supervised Video Anomaly Detection
Giacomo D' Amicantonio, Snehashis Majhi, Quan Kong, Lorenzo Garattoni, Gianpiero Francesca, Francois Bremond, Egor Bondarev
Read the paperCheck your idea against 137,217 papers
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 Temporal Gaussian Splatting loss to the UR-DMU baseline (ablation) | UR-DMU · UCF-Crime | AUC (%): 86.97 -> 87.84 | Improved |
| add class-specific expert models (ablation) | GS-MoE · UCF-Crime | AUC (%): 87.84 -> 89.53 | Improved |
| add gate model combining expert scores with task-aware features (ablation) | GS-MoE · UCF-Crime | AUC (%): 89.53 -> 91.58 | Improved |
| add TGS to the baseline on XD-Violence AUC (ablation) | UR-DMU · XD-Violence | AUC (%): 94.07 -> 94.13 | No clear effect |
| feed task-aware features into the gate model (ablation) | GS-MoE gate · XD-Violence | AP_A (%): 81.45 -> 85.74 | Improved |
| use class experts instead of K-means cluster experts (ablation) | GS-MoE · UCF-Crime | AUC (%): 88.58 -> 91.58 | Improved |
| use gate model instead of soft-MoE (ablation) | GS-MoE · UCF-Crime | AUC (%): 90.14 -> 91.58 | Improved |
| set peak prominence threshold to 0.2 rather than 0.1 or 0.3 (ablation) | TGS · UCF-Crime | AUC (%): 90.34 / 90.75 -> 91.58 | No clear effect |
| mask a class expert at the gate (ablation) | GS-MoE · UCF-Crime per-class | class-wise AUC (%): 86.37 -> 50.02 (Abuse) | Hurt |
| use per-scene experts on UBnormal instead of cluster experts (ablation) | GS-MoE · UBnormal | AUC (%): 69.28 -> 65.95 | Hurt |
| add GS-MoE experts and gate over single UR-DMU model | UR-DMU · UCF-Crime | FPS: 110.09 -> 9.57 | Hurt |
Closest papers by meaning
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.92)
- Weakly Supervised Video Anomaly Detection with Anomaly-Connected Components and Intention Reasoning (CVPR 2026 · similarity 0.91)
- CLUE-VAD: Structured Semantic Clues for Understanding Explainable Events in Video Anomaly Detection (ECCV 2026 · similarity 0.91)
- VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree (NeurIPS 2025 · similarity 0.88)
- Prompt-Enhanced Multiple Instance Learning for Weakly Supervised Video Anomaly Detection (CVPR 2024 · similarity 0.88)
- VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection (AAAI 2024 · similarity 0.88)
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.
