Research map / Label-efficient and robust learning / Recognition and long-tail
DeepKSPD: Learning Kernel-matrix-based SPD Representation for Fine-grained Image Recognition
Melih Engin, Lei Wang, Luping Zhou, Xinwang Liu
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 |
|---|---|---|---|
| replace covariance matrix with kernel matrix in SPD representation (KSPD-VGG vs COV-VGG) (ablation) | VGG-19 features + SVM · Average over 4 datasets | accuracy (%): 77.98 -> 80.1 | Improved |
| replace covariance matrix with kernel matrix in end-to-end network (DeepKSPD vs DeepCOV) (ablation) | end-to-end VGG-19 · Average over 4 datasets | accuracy (%): 81.35 -> 85.13 | Improved |
| add end-to-end learning of descriptors and kernel representation (DeepKSPD vs KSPD-VGG) (ablation) | KSPD · Cars | accuracy (%): 83.5 -> 90.1 | Improved |
| add end-to-end learning (DeepKSPD vs KSPD-VGG) (ablation) | KSPD · MIT Indoor | accuracy (%): 77.2 -> 79.6 | Improved |
| add end-to-end learning (DeepKSPD vs KSPD-VGG) (ablation) | KSPD · Aircraft | accuracy (%): 83.8 -> 86.3 | Improved |
| add end-to-end learning (DeepKSPD vs KSPD-VGG) (ablation) | KSPD · Birds | accuracy (%): 78.5 -> 84.5 | Improved |
| use SPD representation instead of Fisher vector on VGG-19 features (KSPD-VGG vs FV-VGG) | VGG-19 features · Average over 4 datasets | accuracy (%): 73.1 -> 80.1 | Improved |
| use matrix logarithm of covariance (DeepCOV) instead of sign square-root (B-CNN) | end-to-end covariance network · Average over 4 datasets | accuracy (%): 82.5 -> 81.35 | Hurt |
| DeepKSPD vs B-CNN | end-to-end VGG-19 · Cars | accuracy (%): 87.5 -> 90.1 | Improved |
| DeepKSPD vs fine-tuned FV-VGG | VGG-19 · Aircraft | accuracy (%): 78.7 -> 86.3 | Improved |
| DeepKSPD vs fine-tuned FV-VGG | VGG-19 · Birds | accuracy (%): 74.7 -> 84.5 | Improved |
| DeepKSPD vs fine-tuned FV-VGG | VGG-19 · Cars | accuracy (%): 85.7 -> 90.1 | Improved |
Closest papers by meaning
- When VLAD Met Hilbert (CVPR 2016 · similarity 0.88)
- A Riemannian Network for SPD Matrix Learning (AAAI 2017 · similarity 0.87)
- Revisiting Metric Learning for SPD Matrix Based Visual Representation (CVPR 2017 · similarity 0.87)
- Beyond Covariance: SICE and Kernel Based Visual Feature Representation (IJCV 2020 · similarity 0.87)
- Dimensionality Reduction on SPD Manifolds: The Emergence of Geometry-Aware Methods (TPAMI 2017 · similarity 0.86)
- Schur's Positive-Definite Network: Deep Learning in the SPD cone with structure (ICLR 2025 · similarity 0.84)
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.
