Research map / Video / Video segmentation and tracking
TeD-SPAD: Temporal Distinctiveness for Self-Supervised Privacy-Preservation for Video Anomaly Detection
Joseph Fioresi, Ishan Rajendrakumar Dave, Mubarak Shah
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 temporally-distinct triplet loss to the anonymization utility branch (ablation) | SPAct-style anonymizer · VISPR / UCF-Crime | privacy cMAP; AUC: cMAP 52.71 -> 42.21; AUC 73.93 -> 74.81 | Improved |
| add temporal-invariance objective instead of distinctiveness (ablation) | anonymization utility branch · UCF-Crime | AUC (%): 73.93 -> 69.52 | Hurt |
| use LLTC contrastive loss instead of triplet loss (ablation) | temporal distinctiveness loss · UCF-Crime / VISPR | AUC; cMAP: AUC 74.81 -> 75.06; cMAP 42.21 -> 42.44 | No clear effect |
| increase triplet loss weight from 0.1 to 1.0 (ablation) | TeD-SPAD · UCF-Crime / VISPR | AUC; cMAP: AUC 74.81 -> 70.56; cMAP 42.21 -> 55.26 | Hurt |
| lower margin from 1.0 to 0.5 (ablation) | temporal triplet loss · UCF-Crime | AUC (%): 74.81 -> 62.12 | Hurt |
| raise margin from 1.0 to 2.0 (ablation) | temporal triplet loss · UCF-Crime | AUC (%): 74.81 -> 67.30 | Hurt |
| use fixed larger negative clip distance 32 instead of random (ablation) | temporal triplet loss · UCF-Crime / VISPR | cMAP; AUC: cMAP 42.21 -> 28.69; AUC 74.81 -> 70.97 | Mixed |
| use fixed small negative clip distance 8 instead of random (ablation) | temporal triplet loss · UCF-Crime / VISPR | AUC; cMAP: AUC 74.81 -> 76.12; cMAP 42.21 -> 46.34 | Mixed |
| apply TeD-SPAD anonymizer instead of SPAct | privacy-preserving anomaly detection · UCF-Crime / VISPR | privacy cMAP; AUC: cMAP 52.71 -> 42.21; AUC 73.93 -> 74.81 | Improved |
| anonymize XD-Violence videos (vs raw data) | I3D + MGFN · XD-Violence | AP (%): 73.72 -> 60.32 | Hurt |
| use standard Kinetics-pretrained I3D features (no anonymization) | I3D features · VISPR feature-level | privacy cMAP (%): 63.15 vs 52.30 (anonymized) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SPAct — SPAct: Self-Supervised Privacy Preservation for Action Recognition (CVPR 2022) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Collaborative Learning of Anomalies with Privacy (CLAP) for Unsupervised Video Anomaly Detection: A New Baseline (CVPR 2024 · similarity 0.88)
- Federated Weakly Supervised Video Anomaly Detection with Multimodal Prompt (AAAI 2025 · similarity 0.87)
- VPD-100K: Towards Generalizable and Fine-grained Visual Privacy Protection (ICML 2026 · similarity 0.86)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.85)
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.84)
- Just Dance with pi! A Poly-modal Inductor for Weakly-supervised Video Anomaly Detection (CVPR 2025 · similarity 0.83)
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.
