Research map / Label-efficient and robust learning / Noisy labels
Graph Convolutional Label Noise Cleaner: Train a Plug-And-Play Action Classifier for Anomaly Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Alternate-train the TSN Optical Flow classifier with the full GCN cleaner | TSN Optical Flow · UCF-Crime | frame-level AUC (%): 70.87 -> 78.08 | Helped |
| Add indirect supervision (ablation) | Full GCN cleaner · UCF-Crime | frame-level AUC (%): 73.79 -> 74.60 | Helped |
| Add temporal consistency graph convolution (ablation) | GCN cleaner · UCF-Crime | frame-level AUC (%): 70.87 -> 72.93 | Helped |
| Use the temporal graph rather than constant adjacency (ablation) | Temporal consistency module · UCF-Crime | frame-level AUC (%): 67.57 -> 72.93 | Helped |
| Add feature similarity graph convolution (ablation) | GCN cleaner · UCF-Crime | frame-level AUC (%): 70.87 -> 72.44 | Helped |
| Use feature-similarity adjacency rather than constant adjacency (ablation) | Feature similarity module · UCF-Crime | frame-level AUC (%): 67.23 -> 72.44 | Helped |
| Alternate-train the C3D classifier with the GCN cleaner | C3D · ShanghaiTech | frame-level AUC (%): 73.79 -> 76.44 | Helped |
| Continue alternate training from Step-2 to Step-3 (ablation) | TSN Optical Flow · ShanghaiTech | frame-level AUC (%): 84.19 -> 84.13 | Hurt |
| Use alternate training with the GCN cleaner | TSN Gray-scale · UCSD-Peds2 | frame-level AUC (%) | Helped |
Closest papers by meaning
- Learning from Noisy Supervision: A Denoising-Debiasing Framework for Weakly Supervised Video Anomaly Detection (CVPR 2026 · similarity 0.87)
- Exploiting Completeness and Uncertainty of Pseudo Labels for Weakly Supervised Video Anomaly Detection (CVPR 2023 · similarity 0.87)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.86)
- FrameShield: Adversarially Robust Video Anomaly Detection (NeurIPS 2025 · similarity 0.85)
- Rethinking Pseudo-Label Guided Learning for Weakly Supervised Temporal Action Localization from the Perspective of Noise Correction (AAAI 2025 · similarity 0.85)
- Semi-supervised Active Learning for Video Action Detection (AAAI 2024 · similarity 0.85)
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.