Image Classification on ImageNet 1K (val) (Top-1, ReaL, Multilabel)
90.94Top-1 AccGreedy soup
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Greedy soupBackbone=ViT-G/142022.03 | 90.94 | 91.2 | 97.17 | |
| Greedy ensembleBackbone=ViT-G/142022.03 | 90.93 | 91.29 | 97.23 | |
| CoAtNet-7Backbone=CoAtNet-72022.03 | 90.88 | — | — | |
| Best model on each test set (oracle)Backbone=ViT-G/14, Selection=Oracle (each test set)2022.03 | 90.78 | 91.78 | 97.29 | |
| Best model on held out val setBackbone=ViT-G/14, Selection=Held out validation set2022.03 | 90.72 | 91.04 | 96.94 | |
| ViT/G-14Backbone=ViT-G/14, Selection=Reevaluated2022.03 | 90.47 | 90.86 | 96.89 | |
| ViT/G-14Backbone=ViT-G/14, Selection=Original (Zhai et al., 2021)2022.03 | 90.45 | 90.81 | — |