Uncertainty Quantification for Low-resolution Image Classification on CIFAR-100 (test)
82.3Rejection Accuracy @ 90%REPVLM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| REPVLMBackbone=SigLIP ViT-B/16, Pre-training Data=DataComp2026.01 | 82.3 | 0.802 | |
| REPVLMBackbone=SigLIP ViT-B/16, Pre-training Data=Conceptual Caption2026.01 | 81.6 | 0.896 | |
| REPVLMBackbone=SigLIP ViT-B/16, Pre-training Data=LAION2026.01 | 80.9 | 0.721 | |
| ProbVLMBackbone=SigLIP ViT-B/16, Pre-training Data=DataComp2026.01 | 79.8 | 0.069 | |
| ProbVLMBackbone=SigLIP ViT-B/16, Pre-training Data=Conceptual Caption2026.01 | 77.8 | 0.995 | |
| MCDOBackbone=SigLIP ViT-B/16, Pre-training Data=Conceptual Caption2026.01 | 73.9 | 0.722 | |
| MCDOBackbone=SigLIP ViT-B/16, Pre-training Data=DataComp2026.01 | 73.9 | 0.722 | |
| MCDOBackbone=SigLIP ViT-B/16, Pre-training Data=LAION2026.01 | 73.9 | 0.722 | |
| ProbVLMBackbone=SigLIP ViT-B/16, Pre-training Data=LAION2026.01 | 70.7 | -0.137 |