Research map / Video / Action recognition
Temporal-attentive Covariance Pooling Networks for Video 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 |
|---|---|---|---|
| replace global average pooling with plain covariance pooling (ablation) | TSN R50 · Mini-K200 | Top-1 accuracy (%): 71.9 -> 72.1 | No clear effect |
| add matrix power normalization to plain covariance pooling (ablation) | Plain GCP · Mini-K200 | Top-1 accuracy (%): 72.1 -> 74.5 | Helped |
| add temporal-based channel attention (ablation) | Plain GCP + MPN · Mini-K200 | Top-1 accuracy (%): 74.5 -> 75.1 | Helped |
| add temporal-based spatial attention (ablation) | Plain GCP + MPN + f_TCA · Mini-K200 | Top-1 accuracy (%): 75.1 -> 75.6 | Helped |
| add temporal covariance pooling TC-COV with kernel 5 (ablation) | TCP with attention · Mini-K200 | Top-1 accuracy (%): 75.6 -> 76.1 | Helped |
| increase temporal kernel size from 9 to 15 (16 frames) (ablation) | TCPNet · Mini-K200 (16 frames) | Top-1 accuracy (%): 77.7 -> 76.8 | Hurt |
| temporal kernel size 9 instead of 1 (16 frames) (ablation) | TCPNet · Mini-K200 (16 frames) | Top-1 accuracy (%): 77.0 -> 77.7 | Helped |
| use three-frame TSA instead of two-frame TSA (ablation) | TCP · Mini-K200 | Top-1 accuracy (%): 75.5 -> 76.1 | Helped |
| replace Non-local with the proposed TSA (ablation) | TCP attention module · Mini-K200 | Top-1 accuracy (%): +0.5% | Helped |
| replace SE with the proposed TCA (ablation) | TCP attention module · Mini-K200 | Top-1 accuracy (%): +0.5% | Helped |
| replace GAP with TCP | TSN R50 / R152 / X3D-M / TEA · Kinetics-400 | Top-1 accuracy (%): 70.6 -> 75.3; 75.9 -> 78.3; 76.0 -> 77.0; 75.0 -> 76.8 | Helped |
| replace GAP with TCP | TSN, TSM, TPN · Something-Something V1 | Top-1 accuracy (%): TSM 45.6 -> 46.9; TPN 49.0 -> 49.6 | Helped |
Closest papers by meaning
- Revitalizing SVD for Global Covariance Pooling: Halley’s Method to Overcome Over-Flattening (NeurIPS 2025 · similarity 0.87)
- Towards Faster Training of Global Covariance Pooling Networks by Iterative Matrix Square Root Normalization (CVPR 2018 · similarity 0.85)
- What Deep CNNs Benefit From Global Covariance Pooling: An Optimization Perspective (CVPR 2020 · similarity 0.85)
- DropCov: A Simple yet Effective Method for Improving Deep Architectures (NeurIPS 2022 · similarity 0.85)
- Slow and Steady Feature Analysis: Higher Order Temporal Coherence in Video (CVPR 2016 · similarity 0.85)
- TimeBalance: Temporally-Invariant and Temporally-Distinctive Video Representations for Semi-Supervised Action Recognition (CVPR 2023 · similarity 0.85)
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.