Research map / Time series / Foundation models for time series
Non-Linear Temporal Subspace Representations for Activity Recognition
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 |
|---|---|---|---|
| Use kernelized subspace pooling rather than generalized rank pooling | HMDB split-1 FLOW+RGB · HMDB split-1 | accuracy: 68.3 -> 71.9 | Helped |
| Use kernelized subspace pooling rather than generalized rank pooling | MPII split1 FLOW+RGB · MPII split1 | mAP: 53.8 -> 57.2 | Helped |
| Add IDT-FV features to KRP-FS | HMDB (3 splits) · HMDB (3 splits) | average accuracy: 69.8 -> 72.7 | Helped |
| Add IDT-FV features to KRP-FS | MPII Cooking Activities (7 splits) · MPII Cooking Activities (7 splits) | mAP: 70.0 -> 76.1 | Helped |
| Add IDT-FV features to KRP-FS | JHMDB (3 splits) · JHMDB (3 splits) | average accuracy: 73.8 -> 74.2 | Helped |
| Use the improved pre-image formulation with a reconstruction term rather than basic kernelized rank pooling (ablation) | UT-Kinect actions · UT-Kinect actions | accuracy: 84.8 -> 92.1 | Helped |
| Use KRP-FS rather than IBKRP | UT-Kinect actions · UT-Kinect actions | accuracy: 92.1 -> 99.0 | Helped |
| Apply moving-average and signed-square-root preprocessing (ablation) | KRP-FS on HMDB split-1 FLOW · HMDB split-1 FLOW | accuracy: 61.4 -> 61.5 | No clear effect |
| Apply moving-average and signed-square-root preprocessing (ablation) | KRP-FS on HMDB split-1 RGB · HMDB split-1 RGB | accuracy: 51.4 -> 51.6 | No clear effect |
| Remove both moving average and signed-square-root preprocessing (ablation) | KRP-FS on HMDB split-1 FLOW · HMDB split-1 FLOW | accuracy: 61.5 -> 60.8 | Hurt |
| Remove both moving average and signed-square-root preprocessing (ablation) | KRP-FS on HMDB split-1 RGB · HMDB split-1 RGB | accuracy: 51.6 -> 51.3 | Hurt |
| Approximate the KRP-FS kernel with Nyström sampling factor 1/32 rather than 1/2 (ablation) | KRP-FS on HMDB split-1 · HMDB split-1 | accuracy: 61.7 -> 60.56 | Hurt |
Closest papers by meaning
- Generalized Rank Pooling for Activity Recognition (CVPR 2017 · similarity 0.93)
- Discriminative Hierarchical Rank Pooling for Activity Recognition (CVPR 2016 · similarity 0.91)
- Learning End-to-end Video Classification with Rank-Pooling (ICML 2016 · similarity 0.88)
- Dynamic Image Networks for Action Recognition (CVPR 2016 · similarity 0.88)
- Video Representation Learning Using Discriminative Pooling (CVPR 2018 · similarity 0.87)
- Local Temporal Bilinear Pooling for Fine-Grained Action Parsing (CVPR 2019 · similarity 0.86)
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.