Research map / Label-efficient and robust learning / OOD and anomaly detection
RankFeat&RankWeight: Rank-1 Feature/Weight Removal for Out-of-Distribution Detection
Yue Song, Wei Wang, Nicu Sebe
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 RankFeat+RankWeight together | RankFeat (Block 3+4) · ImageNet-1k, ResNetv2-101 | average FPR95: 36.80 -> 16.13 | Improved |
| remove rank-1 matrix instead of keeping only rank-1 (ablation) | RankFeat · ImageNet-1k, ResNetv2-101 | average FPR95: 58.49 -> 39.69 | Improved |
| remove rank-1 matrix instead of rank-2 (ablation) | RankFeat · ImageNet-1k, ResNetv2-101 | average FPR95: 57.92 -> 39.69 | Improved |
| remove rank-1 matrix instead of rank-3 (ablation) | RankFeat · ImageNet-1k, ResNetv2-101 | average FPR95: 61.94 -> 39.69 | Improved |
| apply RankFeat to Block 3 or 4 instead of Block 1 (ablation) | RankFeat · ImageNet-1k, ResNetv2-101 | average FPR95: 76.65 -> 45.80 | Improved |
| add RankWeight to GradNorm | GradNorm · ImageNet-1k, ResNetv2-101 | average FPR95: 54.70 -> 29.99 | Improved |
| add RankWeight to ReAct | ReAct · ImageNet-1k, ResNetv2-101 | average FPR95: 57.66 -> 36.12 | Improved |
| add RankWeight to VRA | VRA · ImageNet-1k, ResNetv2-101 | average FPR95: 30.85 -> 17.75 | Improved |
| add RankWeight to ASH | ASH · ImageNet-1k, ResNetv2-101 | average FPR95: 31.44 -> 15.46 | Improved |
| use power iteration (10 iterations) instead of SVD (ablation) | RankFeat · ImageNet-1k, ResNetv2-101 | average FPR95: 39.69 -> 39.88 | No clear effect |
| use RankWeight alone on ViT-B/16 | ViT-B/16 baseline · ImageNet-1k, ViT-B/16 | average FPR95: 60.75 -> 72.69 | Hurt |
| add RankFeat+RankWeight on Species | ReAct · Species, ResNetv2-101 | average FPR95: 67.02 -> 22.30 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ReAct — ReAct: Out-of-distribution Detection With Rectified Activations (NeurIPS 2021) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SeTAR: Out-of-Distribution Detection with Selective Low-Rank Approximation (NeurIPS 2024 · similarity 0.89)
- A Systematic Analysis of Out-of-Distribution Detection Under Representation and Training Paradigm Shifts (NeurIPS 2026 · similarity 0.87)
- CADRef: Robust Out-of-Distribution Detection via Class-Aware Decoupled Relative Feature Leveraging (CVPR 2025 · similarity 0.87)
- Double Descent Meets Out-of-Distribution Detection: Theoretical Insights and Empirical Analysis on the Role of Model Complexity (NeurIPS 2025 · similarity 0.87)
- Learning to Shape In-distribution Feature Space for Out-of-distribution Detection (NeurIPS 2024 · similarity 0.87)
- Pursuing Feature Separation based on Neural Collapse for Out-of-Distribution Detection (ICLR 2025 · 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.
