Research map / Video / Video segmentation and tracking
Multimodal Motion Conditioned Diffusion Model for Skeleton-based Video Anomaly Detection
Alessandro Flaborea, Luca Collorone, Guido Maria D'Amely di Melendugno, Stefano D'Arrigo, Bardh Prenkaj, Fabio Galasso
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 |
|---|---|---|---|
| use AE-embedding (auxiliary reconstruction of past) for past conditioning instead of input concatenation (ablation) | MoCoDAD · HR-UBnormal | AUC: 65.2 -> 68.4 | Improved |
| use E2E-embedding without reconstruction supervision instead of input concatenation (ablation) | MoCoDAD · HR-UBnormal | AUC: 65.2 -> 64.4 | Hurt |
| use forecasting proxy task instead of in-between imputation (ablation) | MoCoDAD · HR-UBnormal | AUC: 65.7 -> 68.4 | Improved |
| use forecasting instead of random imputation (ablation) | MoCoDAD · HR-UBnormal | AUC: 65.2 -> 68.4 | Improved |
| apply diffusion in latent space (MLD or STS-VAE) instead of pose space (ablation) | MoCoDAD · HR-UBnormal | AUC: 68.4 -> 58.1 (MLD), 62.6 (Latent) | Hurt |
| use 10 diffusion steps instead of 2 (ablation) | MoCoDAD · HR-UBnormal | AUC: 65.0 -> 68.4 | Improved |
| increase diffusion steps from 10 to 50 (ablation) | MoCoDAD · HR-UBnormal | AUC: 68.4 -> 64.4 | Hurt |
| add past-motion conditioning to the diffusion process (ablation) | MoCoDAD · HR-UBnormal | AUC: 55.0 -> 68.4 | Improved |
| use Simplex noise instead of Gaussian corruption (no conditioning) (ablation) | MoCoDAD · HR-UBnormal | AUC: 57.4 -> 53.0 | Hurt |
| increase number of generated futures (quantiles below 0.5 aggregation) (ablation) | MoCoDAD · HR-UBnormal | AUC | Mixed |
| model multiple possible futures with diffusion instead of a deterministic OCC model | skeleton-based VAD · HR-UBnormal | AUC: 65.5 -> 68.4 | Improved |
Closest papers by meaning
- Dual Conditioned Motion Diffusion for Pose-Based Video Anomaly Detection (AAAI 2025 · similarity 0.91)
- Learning Regularity in Skeleton Trajectories for Anomaly Detection in Videos (CVPR 2019 · similarity 0.90)
- Video Anomaly Detection with Motion and Appearance Guided Patch Diffusion Model (AAAI 2025 · similarity 0.89)
- Sequential keypoint density estimator: an overlooked baseline of skeleton-based video anomaly detection (ICCV 2025 · similarity 0.88)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.86)
- Nonisotropic Gaussian Diffusion for Realistic 3D Human Motion Prediction (CVPR 2025 · similarity 0.84)
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.
