Research map / Detection and segmentation / CNN features
Scene Categorization With Spectral Features
Salman H. Khan, Munawar Hayat, Fatih Porikli
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 spectral (DCT) transform layer to Places-VGGnet | Places-VGGnet · MIT-67 | accuracy: 80.9 -> 84.3 | Improved |
| add spectral layer | Places-VGGnet · Places-205 | accuracy: 60.3 -> 61.2 | Improved |
| add spectral layer | Places-VGGnet · SUN-397 | accuracy: 66.9 -> 67.6 | Improved |
| use DFT magnitude instead of DCT (ablation) | spectral layer · MIT-67 | accuracy: 80.1 -> 78.2 | Hurt |
| use DFT phase instead of DCT (ablation) | spectral layer · MIT-67 | accuracy: 80.1 -> 59.6 | Hurt |
| add phase to DFT magnitude (ablation) | DFT magnitude features · MIT-67 | accuracy: 78.2 -> 74.7 | Hurt |
| add spectral features to ImageNet-VGGnet off-the-shelf features | VGGnet + SVM · MIT-67 (domain transfer, with aug.) | accuracy: 71.9 -> 75.7 | Improved |
| add spectral features without data augmentation (ablation) | Places-VGGnet · MIT-67 (no aug.) | accuracy: 77.6 -> 81.7 | Improved |
| replace spectral transform with plain learned FC layer (ablation) | spectral layer · MIT-67 | accuracy: 84.3 -> 81.8 | Hurt |
| replace DCT matrix with random projection (ablation) | spectral layer · MIT-67 | accuracy: 84.3 -> 81.6 | Hurt |
| use RBF kernel SVM instead of linear SVM (ablation) | DCT features · MIT-67 | accuracy: 79.1 -> 80.1 | Improved |
| add spectral layer | GoogleNet · MIT-67 | accuracy: 75.1 -> 77.3 | Improved |
Closest papers by meaning
- Deep Spectral Methods: A Surprisingly Strong Baseline for Unsupervised Semantic Segmentation and Localization (CVPR 2022 · similarity 0.87)
- Affinity CNN: Learning Pixel-Centric Pairwise Relations for Figure/Ground Embedding (CVPR 2016 · similarity 0.86)
- Spectral Feature Transformation for Person Re-Identification (ICCV 2019 · similarity 0.85)
- Modulate Your Spectrum in Self-Supervised Learning (ICLR 2024 · similarity 0.85)
- Deciphering 'What' and 'Where' Visual Pathways from Spectral Clustering of Layer-Distributed Neural Representations (CVPR 2024 · similarity 0.83)
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.83)
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.
