Research map / Architectures and efficiency / CNN design
Deep Scene Image Classification With the MFAFVNet
Yunsheng Li, Mandar Dixit, Nuno Vasconcelos
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 |
|---|---|---|---|
| initialize PCA and MFA layers from pre-trained statistics instead of randomly (ablation) | MFAFVNet (VGG-16, scale 96) · MIT Indoor | accuracy (%): 77.3 -> 80.3 | Improved |
| initialize PCA and MFA layers from pre-trained statistics instead of randomly (ablation) | MFAFVNet (VGG-16, scale 96) · SUN | accuracy (%): 56.2 -> 62.51 | Improved |
| initialize PCA and MFA layers from pre-trained statistics instead of randomly (ablation) | MFAFVNet (AlexNet, scale 96) · MIT Indoor | accuracy (%): 69.82 -> 71.44 | Improved |
| initialize PCA and MFA layers from pre-trained statistics instead of randomly (ablation) | MFAFVNet (AlexNet, scale 96) · SUN | accuracy (%): 50.23 -> 54.14 | Improved |
| increase regularization strength lambda from 0.01 to 1 (ablation) | MFAFVNet (VGG-16) · MIT Indoor | accuracy (%): 79.79 -> 80.3 | Improved |
| increase regularization strength lambda from 0.01 to 1 (ablation) | MFAFVNet (AlexNet) · MIT Indoor | accuracy (%): 70.69 -> 71.44 | Improved |
| increase lambda beyond 1 (10, 100) (ablation) | MFAFVNet (VGG-16) · MIT Indoor | accuracy (%): 80.3 -> 80.12 (lambda 10), 80.14 (lambda 100) | No clear effect |
| combine three patch scales instead of single scale 96 (ablation) | MFAFVNet (VGG-16) · MIT Indoor | accuracy (%): 80.3 -> 81.12 | Improved |
| combine three patch scales instead of single scale 96 (ablation) | MFAFVNet (VGG-19) · MIT Indoor | accuracy (%): 80.5 -> 82.66 | Improved |
| combine three patch scales instead of single scale 96 (ablation) | MFAFVNet (AlexNet) · SUN | accuracy (%): 54.14 -> 57.15 | Improved |
| use larger single patch size 160 instead of 96 (ablation) | MFAFVNet (VGG-16) · MIT Indoor | accuracy (%): 80.3 -> 78.01 | Hurt |
| add first-order term (Eq. 19) to second-order MFA-FV layer (ablation) | MFAFVNet · scene classification | accuracy (%) | No clear effect |
Closest papers by meaning
- Object based Scene Representations using Fisher Scores of Local Subspace Projections (NeurIPS 2016 · similarity 0.88)
- Semantic Fisher Scores for Task Transfer: Using Objects to Classify Scenes (TPAMI 2019 · similarity 0.87)
- RIFD-CNN: Rotation-Invariant and Fisher Discriminative Convolutional Neural Networks for Object Detection (CVPR 2016 · similarity 0.84)
- Modality and Component Aware Feature Fusion For RGB-D Scene Classification (CVPR 2016 · similarity 0.83)
- Learning Transferable Architectures for Scalable Image Recognition (CVPR 2018 · similarity 0.83)
- Training-free Neural Architecture Search through Variance of Knowledge of Deep Network Weights (CVPR 2025 · similarity 0.82)
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.
