Research map / Label-efficient and robust learning / OOD and anomaly detection
Vision-Language Dual-Pattern Matching for Out-of-Distribution Detection
Zihan Zhang, Zhuo Xu, Xiang Xiang
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 visual pattern matching (KL) to textual matching (ablation) | DPM-F · CIFAR-100 ID, LSUN OOD (ViT-B-32) | AUROC: 87.71 -> 95.41 | Improved |
| use KL divergence rather than euclidean distance for visual matching (ablation) | DPM-F · CIFAR-100 ID, CIFAR-10 OOD | AUROC: 61.54 -> 78.72 | Improved |
| add linear projection layers to learnable modules (ablation) | DPM-T · CIFAR-100 (ViT-B-32) | AUROC: 84.99 -> 92.63 | Improved |
| add domain-specific feature aggregation (DSFA) on top of prompt and projection (ablation) | DPM-T · CIFAR-100 (ViT-B-32) | AUROC: 92.63 -> 94.79 | Improved |
| add DSFA alone without prompt or projection (ablation) | DPM-F · CIFAR-100 (ViT-B-32) | AUROC: 84.14 -> 76.93 | Hurt |
| combine L_VM and L_TM losses instead of L_org alone (ablation) | DPM-T · CIFAR-100 (ViT-B-16) | AUROC: 86.65 -> 91.62 | Improved |
| add L_org loss to L_VM + L_TM (ablation) | DPM-T · CIFAR-100 (ViT-B-16) | AUROC: 91.62 -> 91.07 | Hurt |
| use L_TM alone without L_VM (ablation) | DPM-T · CIFAR-100 (ViT-B-16) | AUROC: 61.14 vs 91.62 | Hurt |
| set feature aggregation factor gamma to 0.7 instead of 0 (ablation) | DPM-F · CIFAR-100 (ViT-B-16) | AUROC: about 8% AUROC improvement | Improved |
| increase trainset samples per class (ablation) | DPM-F · CIFAR-100 (ViT-B-16) | mean AUROC: 78.44 -> 85.33 (5 -> 500 per class) | Improved |
| use DPM-F over MCM under covariate shift (ImageNet-A as OOD) | CLIP OOD detection · ImageNet val ID, ImageNet-A OOD | AUROC: 80.74 -> 86.76 | Improved |
| add lightweight training (DPM-T over DPM-F) | DPM · ImageNet-1k, ViT-B-16 | AUROC: 92.78 -> 95.72 | 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 1 of 1 shared result
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
- Overcoming Shortcut Problem in VLM for Robust Out-of-Distribution Detection (CVPR 2025 · similarity 0.88)
- DPU: Dynamic Prototype Updating for Multimodal Out-of-Distribution Detection (CVPR 2025 · similarity 0.88)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.87)
- RUNA: Object-Level Out-of-Distribution Detection via Regional Uncertainty Alignment of Multimodal Representations (AAAI 2025 · similarity 0.87)
- MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities (NeurIPS 2024 · similarity 0.87)
- Envisioning Outlier Exposure by Large Language Models for Out-of-Distribution Detection (ICML 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.
