Research map / Vision-language and multimodal / Cross-modal learning
Modality-Aware Out-of-Distribution Detection for Multi-Modal Action Recognition
Lars Doorenbos, Duc Vu, Serdar Ozsoy, Juergen Gall
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 audio to video and optical flow (ablation) | Ours · EPIC-Kitchens:Kinetics Far-OOD | FPR@95: 12.8 -> 10.8 | Improved |
| add optical flow to video (ablation) | Ours · EPIC-Kitchens:Kinetics Far-OOD | FPR@95: 15.4 -> 12.8 | Improved |
| combine feature, modality and probability scores instead of feature score r(x) alone (ablation) | Ours · Near-OOD | FPR@95: 7.6 | Improved |
| use class-conditional linear map instead of single shared linear map (ablation) | Ours · Kinetics-600 Near-OOD (late fusion) | FPR@95: 63.3 -> 63.0 | No clear effect |
| use class-conditional linear map instead of fully connected map (ablation) | Ours · Kinetics-600 Near-OOD (late fusion) | FPR@95: 63.4 -> 63.0 | No clear effect |
| use full validation set instead of 1% subset (ablation) | Ours · Kinetics-600 Near-OOD | AUC: 77.2 -> 77.0 | No clear effect |
| use probability distributions instead of logits for the distance (ablation) | Ours · Kinetics Near/Far-OOD | FPR@95 | Improved |
| use different distance function (ablation) | Ours · Kinetics Near-OOD and Kinetics:HMDB Far-OOD | FPR@95: 0.6 | No clear effect |
| use invariants (lowest-variance directions) instead of kNN or class-wise Mahalanobis for r(x) (ablation) | Ours · Kinetics Near-OOD | FPR@95 | Improved |
| use relation between uni- and multi-modal predictions instead of A2D uni-modal disagreement (ablation) | Ours s(x) · Kinetics Near-OOD | FPR@95: 9.9 | Improved |
| use relation between uni- and multi-modal predictions instead of A2D uni-modal disagreement (ablation) | Ours s(x) · Kinetics:HMDB Far-OOD | FPR@95: 16.0 | Improved |
| add uni-modal heads post hoc instead of training with them (ablation) | Ours · HMDB Near-OOD | FPR@95: 33.8 -> 34.2 | No clear effect |
Closest papers by meaning
- Leveraging Dark Knowledge for Intrinsic Multimodal Out-of-Distribution Detection (ECCV 2026 · similarity 0.89)
- MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities (NeurIPS 2024 · similarity 0.88)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.87)
- DPU: Dynamic Prototype Updating for Multimodal Out-of-Distribution Detection (CVPR 2025 · similarity 0.86)
- RUNA: Object-Level Out-of-Distribution Detection via Regional Uncertainty Alignment of Multimodal Representations (AAAI 2025 · similarity 0.84)
- ReconBoost: Boosting Can Achieve Modality Reconcilement (ICML 2024 · similarity 0.84)
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.
