Research map / Video / Action recognition
Few-Shot Common Action Localization via Cross-Attentional Fusion of Context and Temporal Dynamics
Juntae Lee, Mihir Jain, Sungrack Yun
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 QtoS Context-CA and Dynamics-CA cross-attention (ablation) | CDC-CA · ActivityNet1.3 single-instance 5-shot | mAP (%): 54.3 -> 66.3 | Improved |
| attend each support video individually in QtoS Context-CA instead of aggregated (ablation) | CDC-CA · ActivityNet1.3 5-shot single-instance | mAP (%): 65.7 -> 66.3 | Improved |
| attend each support video individually in Dynamics-CA instead of aggregated (ablation) | CDC-CA · ActivityNet1.3 5-shot multi-instance | mAP (%): 52.9 -> 53.5 | Improved |
| use individual-to-all instead of individual-to-individual Dynamics-CA (ablation) | CDC-CA · ActivityNet1.3 5-shot single-instance | mAP (%): 65.9 -> 66.3 | Improved |
| apply QtoS Context-CA before Dynamics-CA (keep original order) instead of swapping order (ablation) | CDC-CA · ActivityNet1.3 single-instance 5-shot | mAP (%): 65.6 -> 66.3 | Improved |
| add pairwise ranking loss (ablation) | relational classifier · ActivityNet1.3 single-instance | mAP (%): 61.8 -> 61.9 (1-shot); 65.8 -> 66.0 (5-shot) | Improved |
| add auxiliary relational module with pseudo-class indicator (ablation) | relational classifier · ActivityNet1.3 single-instance | mAP (%): 61.8 -> 61.9 (1-shot); 65.8 -> 66.1 (5-shot) | Improved |
| increase number of pseudo action classes k from 80 to 320 in multi-hot indicator (ablation) | auxiliary relational module · ActivityNet1.3 single-instance 1-shot | mAP (%): 61.9 -> 62.3 | No clear effect |
| increase support videos from 1-shot to 5-shot | CDC-CA · THUMOS14 multi-instance | mAP@0.5: 9.8 -> 15.7 | Improved |
Closest papers by meaning
- Localizing the Common Action Among a Few Videos (ECCV 2020 · similarity 0.91)
- Few-Shot Transformation of Common Actions Into Time and Space (CVPR 2021 · similarity 0.91)
- Learning Temporal Co-Attention Models for Unsupervised Video Action Localization (CVPR 2020 · similarity 0.90)
- Cross-Attentional Audio-Visual Fusion for Weakly-Supervised Action Localization (ICLR 2021 · similarity 0.90)
- Common Action Discovery and Localization in Unconstrained Videos (ICCV 2017 · similarity 0.89)
- Class Semantics-Based Attention for Action Detection (ICCV 2021 · similarity 0.89)
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.
