Research map / Label-efficient and robust learning / OOD and anomaly detection
Rethinking Out-of-Distribution (OOD) Detection: Masked Image Modeling Is All You Need
Jingyao Li, Pengguang Chen, Zexin He, Shaozuo Yu, Shu Liu, Jiaya Jia
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 |
|---|---|---|---|
| Replace classification pretraining with MIM (ablation) | ViT OOD detector · CIFAR-10 → CIFAR-100 OOD | AUROC: 98.6 -> 99.4 | Improved |
| Replace MoCov3 contrastive pretraining with MIM (ablation) | ViT OOD detector · CIFAR-100 → CIFAR-10 OOD | AUROC: 72.8 -> 98.3 | Improved |
| Use MIM rather than MoCov3 pretraining (ablation) | ViT OOD detector · CIFAR-10 one-class | AUROC: 13.3% to 98.66% | Improved |
| Add intermediate fine-tuning after MIM pretraining (ablation) | MOOD · CIFAR-10 one-class | AUROC: 72.2 -> 97.9 | Improved |
| Add intermediate fine-tuning after MIM pretraining (ablation) | MOOD · CIFAR-100 one-class | AUROC: 66.3 -> 96.5 | Improved |
| Add ID fine-tuning after MIM pretraining (ablation) | MOOD · ImageNet-30 one-class | AUROC: 75.2 -> 92.0 | Improved |
| Add both intermediate fine-tuning and ID fine-tuning (ablation) | MOOD · CIFAR-10 multi-class OOD average | AUROC: 74.5 -> 99.7 | Improved |
| Use Mahalanobis distance instead of Softmax for OOD scoring (ablation) | MOOD · CIFAR-10 → SVHN multi-class OOD | AUROC: 88.6 -> 99.8 | Improved |
| Use a single MIM-pretrained ViT instead of R50+ViT (ablation) | OOD detector · CIFAR-100 ID, CIFAR-10 OOD | AUROC: 96.23 -> 98.30% | Improved |
| Use MOOD without OOD exposure instead of R50+ViT trained with 10 OOD samples per class | Near-distribution OOD detector · Near-distribution OOD detection | AUROC: 99.29% -> 99.41% | Improved |
| Use MOOD rather than CSI | One-class OOD detection · CIFAR-10 one-class average | AUROC: +3.5 | Improved |
| Use MOOD rather than SSD+ | Multi-class OOD detection · CIFAR-10 OOD average | AUROC: +2.5 | Improved |
Closest papers by meaning
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 2024 · similarity 0.93)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · similarity 0.90)
- Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox (NeurIPS 2024 · similarity 0.90)
- MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities (NeurIPS 2024 · similarity 0.90)
- Learning to Shape In-distribution Feature Space for Out-of-distribution Detection (NeurIPS 2024 · similarity 0.90)
- ID-like Prompt Learning for Few-Shot Out-of-Distribution Detection (CVPR 2024 · similarity 0.90)
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.
