Research map / Detection and segmentation / Open-set and few-shot detection
UNI-OOD: Unified Object- and Image-level Out-of-Distribution Detection via Cross-Context Attentive Vision-Language Modeling
Yuchuan Li, Azadeh Motamedi, Hyock Ju Kwon, Chul B Park, Il-Min Kim
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 |
|---|---|---|---|
| build background text embedding with inter-text attention instead of comma-separated labels (ablation) | UNI-OOD · Object-level BDD-100k / OpenImages | AUROC: 93.21 -> 98.52 | Improved |
| use inter-text attention instead of averaging individual text embeddings (ablation) | UNI-OOD · Object-level BDD-100k / OpenImages | AUROC: 95.05 -> 98.52 | Improved |
| use inter-text attention instead of MLP combination (ablation) | UNI-OOD · Object-level BDD-100k / OpenImages | AUROC: 97.20 -> 98.52 | Improved |
| add inter-image attention for background (ablation) | UNI-OOD · Object-level BDD-100k / OpenImages | AUROC: 96.02 -> 98.52 | Improved |
| add background text-vision alignment (ablation) | UNI-OOD · Object-level BDD-100k / OpenImages | AUROC: 96.67 -> 98.52 | Improved |
| add intra-image attention for the target object (ablation) | UNI-OOD · Object-level BDD-100k / OpenImages | AUROC: 95.72 -> 98.52 | Improved |
| add target text-vision alignment (ablation) | UNI-OOD · Object-level BDD-100k / OpenImages | AUROC: 96.35 -> 98.52 | Improved |
| add CNNs to the combiner for patch embeddings (ablation) | UNI-OOD · Object-level BDD-100k / OpenImages | AUROC: 97.14 -> 98.52 | Improved |
| add intra-image attention (ablation) | UNI-OOD · Image-level ImageNet-1k / SUN | AUROC: 95.31 -> 97.35 | Improved |
| add inter-image attention (ablation) | UNI-OOD · Image-level ImageNet-1k / iNaturalist | FPR95: 11.86 -> 4.92 | Improved |
| add CNNs to the combiner (ablation) | UNI-OOD · Image-level ImageNet-1k / PLACES | AUROC: 93.75 -> 94.27 | Improved |
| use unified UNI-OOD instead of RUNA | object-level OOD detection · PASCAL-VOC / MS-COCO | AUROC: 92.92 -> 95.07 | 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.
- RUNA — RUNA: Object-Level Out-of-Distribution Detection via Regional Uncertainty Alignment of Multimodal Representations (AAAI 2025) · better on 10 of 10 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
- RUNA: Object-Level Out-of-Distribution Detection via Regional Uncertainty Alignment of Multimodal Representations (AAAI 2025 · similarity 0.92)
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts (CVPR 2025 · similarity 0.90)
- Overcoming Shortcut Problem in VLM for Robust Out-of-Distribution Detection (CVPR 2025 · similarity 0.89)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.89)
- OW-OVD: Unified Open World and Open Vocabulary Object Detection (CVPR 2025 · similarity 0.89)
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.
