Research map / Video / Action recognition
Spot On: Action Localization from Pointly-Supervised Proposals
Pascal Mettes, Jan C. van Gemert, Cees G. M. Snoek
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 |
|---|---|---|---|
| train classifier on best-IoU proposals instead of ground-truth tubes (ablation) | Action localization classifier (APT proposals) · UCF Sports and UCF 101 | localization AP across IoU thresholds: slightly improved accuracy | No clear effect |
| mine proposals from point annotations instead of using ground-truth boxes (ablation) | MIL proposal mining · UCF Sports and UCF 101 | localization AP across IoU thresholds: competitive to the ground truth tubes across all thresholds | No clear effect |
| use point supervision instead of video-level labels only in MIL | Proposal mining · UCF Sports | AUC (IoU 0.2): 0.278 -> 0.545 | Improved |
| use point supervision instead of video-level labels only in MIL | Proposal mining · UCF 101 | mAP (IoU 0.2): 0.136 -> 0.348 | Improved |
| use point supervision instead of video-level labels only in MIL | Proposal mining · Hollywood2Tubes | mAP (IoU 0.2): 0.009 -> 0.143 | Improved |
| point supervision versus box supervision (APT) for training | This work · UCF Sports | AUC (IoU 0.2): 0.546 -> 0.545 | No clear effect |
| point supervision versus box supervision (APT) for training | This work · UCF 101 | mAP (IoU 0.2): 0.345 -> 0.348 | No clear effect |
| annotate points instead of boxes at every frame | Annotation process · UCF Sports and UCF 101 | annotation time: 10-15 times faster | Improved |
| annotate only 10% of frames with points (frame rate 10) (ablation) | Point-based proposal mining · UCF Sports and UCF 101 | mAP at IoU 0.2 and 0.5: 45 times or more annotation speedup | No clear effect |
| annotate fewer frames (lower annotation frame-rate) (ablation) | Point and box supervision · UCF Sports and UCF 101 | mAP: performance generally decreases as less frames are annotated | Hurt |
| iterative proposal mining (10 iterations) guided by points (ablation) | Proposal mining · UCF Sports (qualitative) | IoU of top proposal with ground truth: 0.194 -> 0.627 and 0.401 -> 0.526 | Mixed |
Closest papers by meaning
- Pointly-Supervised Action Localization (IJCV 2018 · similarity 0.96)
- Unsupervised Action Discovery and Localization in Videos (ICCV 2017 · similarity 0.91)
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.90)
- OnPoint: Offline-to-Online Multi-Level Distillation for Point-Supervised Online Temporal Action Localization (ECCV 2026 · similarity 0.90)
- A flexible model for training action localization with varying levels of supervision (NeurIPS 2018 · similarity 0.90)
- HR-Pro: Point-Supervised Temporal Action Localization via Hierarchical Reliability Propagation (AAAI 2024 · similarity 0.90)
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.
