Research map / Label-efficient and robust learning / Recognition and long-tail
Discriminant Feature Extraction by Generalized Difference Subspace
Kazuhiro Fukui, Naoya Sogi, Takumi Kobayashi, Jing‐Hao Xue, Atsuto Maki
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 normalization of projected data to gFDA (ablation) | gFDA · CMU Multi-PIE (128 subjects) | EER (%): 4.65 -> 0.56 | Improved |
| add normalization of projected data to GDS projection (ablation) | GDS projection · CMU Multi-PIE (128 subjects) | EER (%): 8.59 -> 0.43 | Improved |
| add normalization of projected data to GDS projection (ablation) | GDS projection · Yale face B+ (29 classes) | EER (%): 15.72 -> 8.91 | Improved |
| add normalization of projected data to gFDA (ablation) | gFDA · Yale face B+ (29 classes) | EER (%): 11.43 -> 8.28 | Improved |
| add normalization to pcaLDA (ablation) | pcaLDA · CMU Multi-PIE (128 subjects) | recognition rate (%): 97.98 -> 97.90 | No clear effect |
| replace linear gFDA with kernel KgFDA | gFDA · ALOI300 (mean over 2-4 learning images) | mean recognition rate (%): 93.2 -> 93.9 | Improved |
| replace linear gFDA with kernel KgFDA on CNN features | gFDA · ETH80 (CNN features, 5 objects for learning) | recognition rate (%): 82.37 -> 92.08 | Improved |
| apply linear FDA/gFDA/GDS on top of fine-tuned CNN features | ResNet50 · ETH80 (CNN features, 5 objects for learning) | recognition rate (%): 90.81 -> 90.56 | No clear effect |
| apply linear gFDA on top of ResNet18 features | ResNet18 · MNIST (CNN features) | recognition rate (%): 99.17 -> 98.83 | Hurt |
| apply kernel KGDS on top of ResNet18 features | ResNet18 · MNIST (CNN features) | recognition rate (%): 99.17 -> 99.61 | Improved |
| apply linear gFDA on top of ResNet50 features | ResNet50 · CIFAR10 (CNN features) | recognition rate (%): 93.21 -> 91.84 | Hurt |
| apply kernel KGDS on top of ResNet50 features | ResNet50 · CIFAR10 (CNN features) | recognition rate (%): 93.21 -> 93.92 | Improved |
Closest papers by meaning
- The Geometry of Nonlinear Embeddings in Kernel Discriminant Analysis (TPAMI 2022 · similarity 0.87)
- Neural Fisher Discriminant Analysis: Optimal Neural Network Embeddings in Polynomial Time (ICML 2022 · similarity 0.86)
- A Generalized Geometric Theoretical Framework of Centroid Discriminant Analysis for Linear Classification of Multi-dimensional Data (ICLR 2026 · similarity 0.86)
- Local Feature Discriminant Projection (TPAMI 2016 · similarity 0.85)
- Separability-Oriented Subclass Discriminant Analysis (TPAMI 2017 · similarity 0.85)
- Trace Ratio Optimization With Feature Correlation Mining for Multiclass Discriminant Analysis (AAAI 2018 · 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.
