Research map / Vision-language and multimodal / CLIP and zero-shot
Respecting Modality Gap in Post-hoc Out-of-distribution Detection with Pre-trained Vision-Language Models
Yuanwei Hu, Bo Peng, Yadan Luo, Zhen Fang, Ling Chen, Jie Lu
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 |
|---|---|---|---|
| learn visual-space prototypes online instead of text-as-prototype with adaptive negative cache (vs AdaNeg+NegLabel) | NegLabel pipeline on CLIP ViT-B/16 · ImageNet-1K, 4 OOD sets average | FPR95: 18.92 -> 12.79 | Improved |
| online visual prototype learning vs AdaNeg+NegLabel on CIFAR-100 Near-OOD | NegLabel pipeline · OpenOOD CIFAR-100 ID, Near-OOD | FPR95: 59.07 -> 45.70 | Improved |
| online visual prototype learning vs AdaNeg+NegLabel on ImageNet Near-OOD | NegLabel pipeline · OpenOOD ImageNet-1K ID, Near-OOD | FPR95: 67.51 -> 52.86 | Improved |
| swap visual encoder from ResNet50 to ViT-B/32 to ViT-L/14 (ablation) | Ours · ImageNet-1K, 4 OOD sets average | FPR95: 15.88 -> 15.11 -> 12.12 | Improved |
| increase input resolution to 336x336 (ablation) | Ours with ViT-L/14 · ImageNet-1K, 4 OOD sets average | FPR95: 12.12 -> 11.21 | Improved |
| vary kappa, rho and beta hyper-parameters (ablation) | Ours · ImageNet-1K | FPR95 | No clear effect |
| use LAPT learned prompts instead of pre-defined prompt (ablation) | Ours · ImageNet-1K | FPR95: 12.79 -> 10.16 | Improved |
| apply method under domain-shifted ID data (ImageNet-A) | Ours · ImageNet-A as ID | FPR95: 35.30 -> 20.99 | Improved |
| online visual prototypes on ImageNet-R iNaturalist subset | Ours · ImageNet-R as ID, iNaturalist | FPR95: 0.89 -> 1.12 | Hurt |
Closest papers by meaning
- Negative Label Guided OOD Detection with Pretrained Vision-Language Models (ICLR 2024 · similarity 0.89)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.89)
- CRoFT: Robust Fine-Tuning with Concurrent Optimization for OOD Generalization and Open-Set OOD Detection (ICML 2024 · similarity 0.88)
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Overcoming the Pitfalls of Vision-Language Model Finetuning for OOD Generalization (ICLR 2024 · similarity 0.88)
- DPU: Dynamic Prototype Updating for Multimodal Out-of-Distribution Detection (CVPR 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.
