Image Classification on TinyImgNet long-tailed (test)
35Avg RecallLA with class priors
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LA with class priorsBackbone=ResNet-18, Objective=Maximizing worst-case recall2021.07 | 35 | 9.5 | |
| CSL [Logit-adjusted]Backbone=ResNet-18, Objective=Maximizing worst-case recall2021.07 | 32.6 | 31 | |
| ERMBackbone=ResNet-18, Objective=Maximizing worst-case recall2021.07 | 31.9 | 1 | |
| CSL [Re-weighted]Backbone=ResNet-18, Objective=Maximizing worst-case recall2021.07 | 30.6 | 9.8 |