Certified Selective Classification on ImageNet backbones V2 (val)
82.28Top-1 AccuracySCORC
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SCORCModel=ResNet-152 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 82.28 | 82.9 | 82.9 | 0 | 33 | |
| A(πmin) (textbook Hoeffding–CRC)Model=ResNet-152 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 82.28 | 0 | — | — | — | |
| A(pLCB) (sophisticated Hoeffding)Model=ResNet-152 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 82.28 | 82.9 | — | — | — | |
| SCORCModel=ResNet-101 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 81.91 | 80.5 | 80.5 | 0 | 33 | |
| A(πmin) (textbook Hoeffding–CRC)Model=ResNet-101 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 81.91 | 0 | — | — | — | |
| A(pLCB) (sophisticated Hoeffding)Model=ResNet-101 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 81.91 | 80.5 | — | — | — | |
| SCORCModel=ResNet-50 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 80.84 | 27.5 | 27.5 | 0 | 21 | |
| A(πmin) (textbook Hoeffding–CRC)Model=ResNet-50 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 80.84 | 0 | — | — | — | |
| A(pLCB) (sophisticated Hoeffding)Model=ResNet-50 V2, alpha=0.05, pi_min=0.01, delta=0.05, ncert=330002026.06 | 80.84 | 27.5 | — | — | — |