Research map / Label-efficient and robust learning / Recognition and long-tail
Open-Set Recognition: A Good Closed-Set Classifier is All You Need
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 maximum logit score instead of maximum softmax probability (ablation) | VGG32 OSR baseline · TinyImageNet | AUROC: 68.9 -> 69.6 | Helped |
| Train for 200 epochs with cosine schedule instead of the 100-epoch step schedule (ablation) | VGG32 MLS baseline · TinyImageNet | AUROC: 69.6 -> 74.8 | Helped |
| Replace RandCrop with CutOut (ablation) | VGG32 MLS baseline · TinyImageNet | AUROC: 74.8 -> 75.4 | Helped |
| Replace RandCrop with RandAug (ablation) | VGG32 MLS baseline · TinyImageNet | AUROC: 74.8 -> 76.6 | Helped |
| Extend training from 200 to 600 epochs and use cosine schedule with 2 restarts (ablation) | VGG32 MLS baseline · TinyImageNet | AUROC: 76.6 -> 78.2 | Helped |
| Add learning-rate warmup (ablation) | VGG32 MLS baseline · TinyImageNet | AUROC: 78.2 -> 78.4 | Helped |
| Add label smoothing (ablation) | VGG32 MLS baseline · TinyImageNet | AUROC: 78.4 -> 83.0 | Helped |
| Add an ensemble (ablation) | VGG32 MLS baseline · TinyImageNet | AUROC: 83.0 -> 84.0 | Helped |
| Use maximum logit score rather than maximum softmax probability | OSR baseline · Standard OSR benchmarks | AUROC: +1.5 MNIST; +8.5 SVHN; +25.9 CIFAR10; +16.3 CIFAR + 10; +16.0 CIFAR + 50; +25.3 TinyImageNet | Helped |
| Apply stronger training strategies to ARPL + CS | ARPL + CS · CIFAR + 10 | AUROC: 97.1 -> 98.1 | Helped |
| Apply stronger training strategies to ARPL + CS | ARPL + CS · TinyImageNet | AUROC: 78.2 -> 82.5 | Helped |
| Evaluate on the semantically harder ImageNet split rather than the Easy split | MLS and ARPL+ · ImageNet | AUROC: 5-6% worse AUROC | Hurt |
Closest papers by meaning
- From Coarse to Fine-Grained Open-Set Recognition (CVPR 2024 · similarity 0.91)
- All Beings Are Equal in Open Set Recognition (AAAI 2024 · similarity 0.89)
- Unlocking Better Closed-Set Alignment Based on Neural Collapse for Open-Set Recognition (AAAI 2025 · similarity 0.89)
- Frequency Shuffling and Enhancement for Open Set Recognition (AAAI 2024 · similarity 0.88)
- Exploring Diverse Representations for Open Set Recognition (AAAI 2024 · similarity 0.88)
- Unlocking the Power of Open Set: A New Perspective for Open-Set Noisy Label Learning (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.