Image Classification on CIFAR100 (Accuracy, ECE, NLL, Brier)
88.2AccuracyLoRA-Ensemble
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LoRA-EnsembleBackbone=DINO ViT-S/14, Evaluation Protocol=Fine-tuning, Training Duration=10 epochs, Number of Ensemble Members=42026.01 | 88.2 | 1.1 | 0.37 | 0.17 | |
| SV-EnsembleBackbone=DINO ViT-S/14, Evaluation Protocol=Fine-tuning, Training Duration=10 epochs, Number of Ensemble Members=42026.01 | 85.6 | 2.1 | 0.47 | 0.21 | |
| Deep EnsembleBackbone=DINO ViT-S/14, Evaluation Protocol=Fine-tuning, Training Duration=10 epochs, Number of Ensemble Members=42026.01 | 85.4 | 3.9 | 0.5 | 0.21 | |
| Batch EnsembleBackbone=DINO ViT-S/14, Evaluation Protocol=Fine-tuning, Training Duration=10 epochs, Number of Ensemble Members=42026.01 | 82.9 | 2.9 | 0.57 | 0.24 | |
| Single w/ SVFBackbone=DINO ViT-S/14, Evaluation Protocol=Fine-tuning, Training Duration=10 epochs2026.01 | 82.6 | 1.1 | 0.57 | 0.25 | |
| SingleBackbone=DINO ViT-S/14, Evaluation Protocol=Fine-tuning, Training Duration=10 epochs2026.01 | 82.1 | 6.2 | 0.65 | 0.26 | |
| MC DropoutBackbone=DINO ViT-S/14, Evaluation Protocol=Fine-tuning, Training Duration=10 epochs2026.01 | 78.6 | 5.3 | 0.74 | 0.3 |