Research map / Label-efficient and robust learning / Recognition and long-tail
Hybrid Models for Open Set Recognition
Hongjie Zhang, Ang Li, Jie Guo, Yanwen Guo
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 |
|---|---|---|---|
| train encoder jointly with classifier and flow instead of pretrained frozen encoder (ablation) | OpenHybrid · MNIST/SVHN/CIFAR10 | F-score: 0.847 -> 0.942 (MNIST), 0.842 -> 0.912 (SVHN), 0.791 -> 0.865 (CIFAR10) | Improved |
| joint training instead of pretrained encoder (ablation) | OpenHybrid · TinyImageNet | F-score: 0.674 -> 0.753 | Improved |
| joint training instead of pretrained encoder (ablation) | OpenHybrid · CIFAR+10 / CIFAR+50 | F-score: 0.783 -> 0.903 (CIFAR+10), 0.761 -> 0.888 (CIFAR+50) | Improved |
| learn embedding jointly with flow density estimator instead of flow on raw-embedding baselines (DIGLM) | OpenHybrid · CIFAR10 | AUROC: 0.583 -> 0.950 | Improved |
| add outlier exposure to DIGLM | DIGLM · CIFAR10 | AUROC: 0.583 -> 0.655 | Improved |
| OpenHybrid versus DIGLM+OE in F-score | OpenHybrid · CIFAR10 | F-score: 0.701 -> 0.865 | Improved |
| increase trade-off lambda from 0.5 to 2 (ablation) | OpenHybrid · MNIST | AUROC: 0.993, 0.995, 0.998 for lambda 0.5, 1, 2 | No clear effect |
| increase number of residual blocks from 4 to 16 (ablation) | OpenHybrid · CIFAR10 | AUROC: 0.945, 0.948, 0.950, 0.958 for 4, 8, 10, 16 blocks | No clear effect |
| compare against outlier exposure on cross-dataset OOD CIFAR10 -> CIFAR100 | OpenHybrid · CIFAR10 -> CIFAR100 | AUROC: 0.933 -> 0.951 | Improved |
| compare against outlier exposure on CIFAR100 -> CIFAR10 | OpenHybrid · CIFAR100 -> CIFAR10 | AUROC: 0.757 -> 0.856 | Improved |
| use full CIFAR10 as known and SVHN as unknown | OpenHybrid · CIFAR10 vs SVHN | AUROC | Improved |
Closest papers by meaning
- Generative-Discriminative Feature Representations for Open-Set Recognition (CVPR 2020 · similarity 0.91)
- Open-Set Likelihood Maximization for Few-Shot Learning (CVPR 2023 · similarity 0.89)
- OpenGAN: Open-Set Recognition via Open Data Generation (ICCV 2021 · similarity 0.89)
- Open-Set Recognition: A Good Closed-Set Classifier is All You Need (ICLR 2022 · similarity 0.89)
- Exploring Diverse Representations for Open Set Recognition (AAAI 2024 · similarity 0.88)
- All Beings Are Equal in Open Set Recognition (AAAI 2024 · 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.
