Research map / Label-efficient and robust learning / Recognition and long-tail
On the Eigenvalues of Global Covariance Pooling for Fine-grained Visual Recognition
Yue Song, Nicu Sebe, Wei Wang
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 SEB to MPN-COV | MPN-COV (VGG-16) · Birds/Aircrafts/Cars | accuracy: 83.4 -> 85.7 (Birds), 89.2 -> 90.7, 90.9 -> 93.0 | Improved |
| add SEB to MPN-COV | MPN-COV (ResNet-50) · Birds/Aircrafts/Cars | accuracy: 84.3 -> 86.2, 89.9 -> 91.4, 91.7 -> 93.6 | Improved |
| add SEB to MPN-COV | MPN-COV (EfficientNet-b5) · Birds/Aircrafts/Cars | accuracy: 87.3 -> 88.2, 92.4 -> 93.5, 93.4 -> 94.6 | Improved |
| add SEB to MPN-COV | MPN-COV (Inception V3) · Birds/Aircrafts/Cars | accuracy: 82.7 -> 84.8, 89.9 -> 90.9, 90.6 -> 92.5 | Improved |
| add SEB to MPN-COV | MPN-COV (DenseNet-169) · Birds/Aircrafts/Cars | accuracy: 86.3 -> 87.6, 90.8 -> 92.3, 92.9 -> 94.0 | Improved |
| add SEB to MPN-COV | MPN-COV (EfficientNet-b5) · Cars/Dogs/iNat | accuracy: 93.4 -> 94.6, 92.1 -> 93.0, 70.8 -> 72.3 | Improved |
| add SEB to MPN-COV | MPN-COV (AlexNet) · ImageNet val | top-1 accuracy: 51.23 -> 56.68 | Improved |
| add SEB to MPN-COV | MPN-COV (ResNet-50) · ImageNet val | top-1 accuracy: 77.07 -> 75.79 | Hurt |
| add the plain Q path to the SEB path (ablation) | SEB · Birds/Aircrafts/Cars (ResNet-50) | accuracy: 85.6 -> 86.2, 91.1 -> 91.4, 93.0 -> 93.6 | Improved |
| use SEB path only instead of plain MPN-COV (ablation) | MPN-COV · Birds/Aircrafts/Cars (ResNet-50) | accuracy: 84.3 -> 85.6, 89.9 -> 91.1, 91.7 -> 93.0 | Improved |
| use ||QS^T||_F instead of ||Q||_F as scaling factor (ablation) | SEB · Birds/Aircrafts/Cars (ResNet-50) | accuracy: 83.8 -> 86.2, 90.3 -> 91.4, 92.1 -> 93.6 | Improved |
| use constant 100 as scaling factor instead of ||QS^T||_F (ablation) | SEB · Birds/Aircrafts/Cars (ResNet-50) | accuracy: 86.2 -> 82.4, 91.4 -> 88.5, 93.6 -> 89.3 | Hurt |
Closest papers by meaning
- Revitalizing SVD for Global Covariance Pooling: Halley’s Method to Overcome Over-Flattening (NeurIPS 2025 · similarity 0.89)
- Deep CNNs Meet Global Covariance Pooling: Better Representation and Generalization (TPAMI 2020 · similarity 0.89)
- Towards a Deeper Understanding of Global Covariance Pooling in Deep Learning: An Optimization Perspective (TPAMI 2023 · similarity 0.87)
- Temporal-attentive Covariance Pooling Networks for Video Recognition (NeurIPS 2021 · similarity 0.87)
- DropCov: A Simple yet Effective Method for Improving Deep Architectures (NeurIPS 2022 · similarity 0.87)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · 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.
