Research map / Video / Action recognition
PointTAD: Multi-Label Temporal Action Detection with Learnable Query Points
Read the paperCheck 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 |
|---|---|---|---|
| replace segment-based RoI-align features with learnable query points (ablation) | PointTAD · MultiTHUMOS | detection-mAP | Helped |
| use deformable convolution instead of direct interpolation at query points (ablation) | PointTAD · MultiTHUMOS | detection-mAP | Helped |
| add frame mixing to channel-only mixing (parallel) (ablation) | PointTAD · MultiTHUMOS | avg detection-mAP | Helped |
| cascade frame->channel mixing instead of parallel/single (ablation) | PointTAD · MultiTHUMOS | avg detection-mAP: -1% avg-mAP | Hurt |
| cascade channel->frame mixing instead of single (ablation) | PointTAD · MultiTHUMOS | avg detection-mAP: -3% avg-mAP | Hurt |
| partial min-max vs min-max transformation (ablation) | PointTAD · MultiTHUMOS | avg detection-mAP | No clear effect |
| vary query point initialization (uniform, normal, constant) (ablation) | PointTAD · MultiTHUMOS | avg detection-mAP | No clear effect |
| increase number of query points N_s up to 21 (ablation) | PointTAD · MultiTHUMOS | detection-mAP | Helped |
| increase input resolution from 96^2 to 128^2 (ablation) | PointTAD · MultiTHUMOS | avg detection-mAP: +3.9% absolute | Helped |
| increase input resolution from 128^2 to 160^2 (ablation) | PointTAD · MultiTHUMOS | avg detection-mAP: +1% absolute | Helped |
| increase input resolution from 160^2 to 192^2 (ablation) | PointTAD · MultiTHUMOS | avg detection-mAP: +0.6% absolute | Helped |
| scale offsets by action duration instead of clip duration (ablation) | PointTAD · MultiTHUMOS | detection-mAP | Helped |
Closest papers by meaning
- Dual DETRs for Multi-Label Temporal Action Detection (CVPR 2024 · similarity 0.92)
- Modeling Multi-Label Action Dependencies for Temporal Action Localization (CVPR 2021 · similarity 0.90)
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.88)
- HR-Pro: Point-Supervised Temporal Action Localization via Hierarchical Reliability Propagation (AAAI 2024 · similarity 0.88)
- Action-Agnostic Point-Level Supervision for Temporal Action Detection (AAAI 2025 · similarity 0.88)
- Text-Infused Attention and Foreground-Aware Modeling for Zero-Shot Temporal Action Detection (NeurIPS 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.