Research map / Label-efficient and robust learning / Recognition and long-tail
The Devil is in the Wrongly-classified Samples: Towards Unified Open-set Recognition
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the existing uncertainty scores to reject InW samples together with OoD samples | OSR methods · CIFAR100 InD / TinyImageNet OoD | AUROC: 84.69 -> 84.90 | Helped |
| Add temperature scaling, T=2 (ablation) | SoftMax · Temperature scaling, w/ pre-training | UOSR AUROC: 90.51 -> 92.40 | Helped |
| Increase temperature from T=5 to T=20 (ablation) | SoftMax · Temperature scaling, w/ pre-training | OSR AUROC: 90.86 -> 91.07 | Helped |
| Increase temperature from T=5 to T=20 (ablation) | SoftMax · Temperature scaling, w/ pre-training | UOSR AUROC: 92.96 -> 92.92 | No clear effect |
| Use reference OoD samples at evaluation | few-shot UOSR · Image domain | AUROC: 93.45 -> 95.51 | Helped |
| Use reference OoD samples at evaluation | few-shot UOSR · Video domain | AUROC: 93.38 -> 96.19 | Helped |
| Add real outlier exposure data (ablation) | SoftMax · CIFAR100 InD / TinyImageNet OoD, w/ pre-training | UOSR AUROC: 90.50 -> 92.54 | Helped |
| Use real rather than generated outlier data (ablation) | PROSER · CIFAR100 InD / TinyImageNet OoD | InC/OoD AUROC: 87.04 -> 88.36 | Helped |
| Use real rather than generated outlier data (ablation) | VOS · CIFAR100 InD / TinyImageNet OoD | InC/OoD AUROC: 87.51 -> 91.44 | Helped |
| Add outlier data through ARPL+CS rather than ARPL (ablation) | ARPL · CIFAR100 InD / TinyImageNet OoD, w/o pre-training | UOSR AUROC: 88.13 -> 87.65 | Hurt |
| Use NGC to identify/correct noisy outlier data (ablation) | model trained with noisy outliers · Noisy outlier data | closed-set accuracy | Helped |
| Increase outlier-noise level (ablation) | model trained without NGC · Noisy outlier data | closed-set accuracy | Hurt |
Closest papers by meaning
- All Beings Are Equal in Open Set Recognition (AAAI 2024 · similarity 0.88)
- Meta Evidential Transformer for Few-Shot Open-Set Recognition (ICML 2024 · similarity 0.87)
- From Coarse to Fine-Grained Open-Set Recognition (CVPR 2024 · similarity 0.87)
- Frequency Shuffling and Enhancement for Open Set Recognition (AAAI 2024 · similarity 0.86)
- How Does Unlabeled Data Provably Help Out-of-Distribution Detection? (ICLR 2024 · similarity 0.85)
- Out-of-Distribution Detection in Long-Tailed Recognition with Calibrated Outlier Class Learning (AAAI 2024 · 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.