Research map / Label-efficient and robust learning / OOD and anomaly detection
OOD-CV-v2 : An Extended Benchmark for Robustness to Out-of-Distribution Shifts of Individual Nuisances in Natural Images
Bingchen Zhao, Jiahao Wang, Wufei Ma, Artur Jesslen, Siwei Yang, Shaozuo Yu, Oliver Zendel, Christian Theobalt, Alan Yuille, Adam Kortylewski
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 |
|---|---|---|---|
| shift test data from i.i.d. to OOD occlusion | ResNet-50 classification · OOD-CV-v2 | top-1 accuracy: 83.9% -> 58.5% (-15.4% for occlusion vs -7.8% for pose) | Hurt |
| shift test data from i.i.d. to OOD context | Faster-RCNN · OOD-CV-v2 | mAP: 40.6% -> 27.0% | Hurt |
| add AugMix / PixMix / style transfer augmentation for appearance nuisances (ablation) | ResNet-50 classification · OOD-CV-v2 weather | top-1 accuracy: 71.5% -> 74.5% (AugMix), 75.6% (PixMix) | Improved |
| add style transfer augmentation (ablation) | ResNet-50 classification · OOD-CV-v2 shape | top-1 accuracy: 68.7% -> 65.6% | Hurt |
| add style transfer augmentation (ablation) | Faster-RCNN · OOD-CV-v2 context | mAP: 27.0% -> 31.1% | Improved |
| add PixMix augmentation (ablation) | Faster-RCNN · OOD-CV-v2 occlusion | mAP: 10.0% -> 18.5% | Improved |
| add AugMix augmentation (ablation) | Res50-Specific pose estimation · OOD-CV-v2 pose | Acc@pi/6: 45.2% -> 44.8% | No clear effect |
| replace ResNet-50 backbone with Swin-T transformer | classification, detection, pose · OOD-CV-v2 | OOD robustness | No clear effect |
| use smaller MobileNetV3-L with AugMix instead of ResNet-50 | classification · OOD-CV-v2 context | top-1 accuracy gain from AugMix: +3.9% (ResNet-50) vs +0.9% (MobileNetV3) | Hurt |
| use two-stage Faster-RCNN instead of one-stage RetinaNet | object detection · OOD-CV-v2 | mAP i.i.d.: 43.9% -> 40.6% | Hurt |
| use NeMo with explicit 3D geometry instead of Res50-Specific | 3D pose estimation · OOD-CV-v2 pose | Acc@pi/6: 45.2% -> 56.9% | Improved |
| add AugMix to NeMo (ablation) | NeMo · OOD-CV-v2 texture | Acc@pi/6: 52.6% -> 57.8% | Improved |
Closest papers by meaning
- A Bayesian Approach to OOD Robustness in Image Classification (CVPR 2024 · similarity 0.90)
- LAION-C: An Out-of-Distribution Benchmark for Web-Scale Vision Models (ICML 2025 · similarity 0.89)
- Unexplored Faces of Robustness and Out-of-Distribution: Covariate Shifts in Environment and Sensor Domains (CVPR 2024 · similarity 0.89)
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · similarity 0.87)
- COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts (CVPR 2025 · similarity 0.87)
- OODRobustBench: a Benchmark and Large-Scale Analysis of Adversarial Robustness under Distribution Shift (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.
