Research map / Label-efficient and robust learning / OOD and anomaly detection
Utilizing Patch-level Category Activation Patterns for Multiple Class Novelty Detection
Poojan Oza, Vishal M. Patel
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 patch-level mismatch novelty network on top of global network (ablation) | VGG16 global network · Overall (4 datasets, VGG16) | AUC: 0.841 -> 0.883 | Improved |
| add global network to local-only network (ablation) | BagNet local network · Overall (4 datasets, VGG16) | AUC: 0.739 -> 0.883 | Improved |
| replace naive Global+Local score fusion with mismatch-trained novelty network (ablation) | Global+Local · Overall (4 datasets, VGG16) | AUC: 0.837 -> 0.883 | Improved |
| naive fusion of global and local scores (no mismatch training) (ablation) | Global Only · Overall (4 datasets, VGG16) | AUC: 0.841 -> 0.837 | No clear effect |
| add mismatch-trained novelty network (ablation) | Global Only VGG16 · Stanford Dogs | AUC: 0.766 -> 0.827 | Improved |
| add reference dataset regularization on penultimate activations | Proposed · Overall (4 datasets, VGG16 and AlexNet avg) | AUC: 0.857 -> 0.879 | Improved |
| add reference dataset regularization | Proposed · Stanford Dogs VGG16 | AUC: 0.827 -> 0.873 | Improved |
| add reference dataset regularization | Proposed · FounderType-200 AlexNet | AUC: 0.798 -> 0.801 | No clear effect |
| use proposed method without reference data instead of DTMND without reference data | DTMND · Overall (4 datasets) | AUC: 0.80 -> 0.857 | Improved |
| use reference dataset via simple fine-tuning | Fine-tune · Overall (4 datasets) | AUC: 0.801 -> 0.800 | No clear effect |
| replace Fine-tune score with OpenMax | Fine-tune · Overall (4 datasets) | AUC: 0.801 -> 0.809 | Improved |
| use proposed method versus Fine-tune baseline | Fine-tune · FounderType-200 VGG16 | AUC: 0.841 -> 0.876 | Mixed |
Closest papers by meaning
- Hierarchical Novelty Detection for Visual Object Recognition (CVPR 2018 · similarity 0.89)
- Deep Transfer Learning for Multiple Class Novelty Detection (CVPR 2019 · similarity 0.88)
- OCGAN: One-Class Novelty Detection Using GANs With Constrained Latent Representations (CVPR 2019 · similarity 0.87)
- Adversarially Learned One-Class Classifier for Novelty Detection (CVPR 2018 · similarity 0.87)
- Multiple Class Novelty Detection Under Data Distribution Shift (ECCV 2020 · similarity 0.87)
- Learning Deep Classifiers Consistent With Fine-Grained Novelty Detection (CVPR 2021 · similarity 0.87)
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.
