Research map / Video / Action recognition
Rank Pooling for Action Recognition
Basura Fernando, Efstratios Gavves, Jose Oramas M., Amir Ghodrati, Tinne Tuytelaars
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 |
|---|---|---|---|
| use time varying mean vectors instead of independent frames (ablation) | FDRP with Fisher vectors · Hollywood2 | mAP: 57.4 -> 63.6 | Improved |
| use moving average (T=50) instead of independent frames (ablation) | FDRP · Hollywood2 | mAP: 57.4 -> 60.8 | Improved |
| add reverse direction ranking (RFDRP vs FDRP) (ablation) | rank pooling · Hollywood2 | mAP: 63.6 -> 67.9 | Improved |
| add non-linear feature map (NL-RFDRP vs RFDRP) (ablation) | rank pooling · Hollywood2 | mAP: 67.9 -> 69.6 | Improved |
| add non-linear map to forward rank pooling (ablation) | FDRP · MPII Cooking | mAP: 62.4 -> 65.6 | Improved |
| add non-linear map to reverse+forward rank pooling (ablation) | RFDRP · MPII Cooking | mAP: 63.5 -> 62.7 | Hurt |
| combine local features with NL-RFDRP | Local features · Hollywood2 | mAP: 62.9 -> 70.0 | Improved |
| combine local features with NL-RFDRP | Local features · MPII Cooking | mAP: 63.4 -> 72.0 | Improved |
| use time varying mean vectors before average pooling | average pooling · Hollywood2 MBH | mAP: +0.5% | No clear effect |
| use rank pooling (SVR) instead of PCA subspace pooling (ablation) | parametric pooling · Hollywood2 MBH | mAP: 56.4 -> 66.5 | Improved |
| replace rank pooling with neural network pooling (ablation) | rank pooling · Hollywood2 MBH | mAP: 66.5 -> 21.1 | Hurt |
| posneg feature map for ranking and classification (ablation) | rank pooling Fisher vectors · Hollywood2 MBH | mAP: 54.0 -> 66.1 | Improved |
Closest papers by meaning
- Discriminatively Learned Hierarchical Rank Pooling Networks (IJCV 2017 · similarity 0.95)
- Discriminative Hierarchical Rank Pooling for Activity Recognition (CVPR 2016 · similarity 0.93)
- Second-order Temporal Pooling for Action Recognition (IJCV 2018 · similarity 0.92)
- Learning End-to-end Video Classification with Rank-Pooling (ICML 2016 · similarity 0.92)
- Generalized Rank Pooling for Activity Recognition (CVPR 2017 · similarity 0.92)
- Action Recognition with Dynamic Image Networks (TPAMI 2017 · similarity 0.91)
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.
