Discrete Representations Strengthen Vision Transformer Robustness

Open paper
Benchmark Model Rank Results
domain-generalization-on-imagenet-cDiscreteViT (Im21k)#11mean Corruption Error (mCE): 38.74Number of params: 87M
domain-generalization-on-imagenet-cDrViT#21mean Corruption Error (mCE): 46.22
domain-generalization-on-imagenet-cDiscreteViT#22mean Corruption Error (mCE): 46.22Number of params: 87M
domain-generalization-on-imagenet-rDiscreteViT#19Top-1 Error Rate: 44.74
domain-generalization-on-imagenet-sketchDrViT#15Top-1 accuracy: 44.72
domain-generalization-on-stylized-imagenetDiscreteViT#3Top 1 Accuracy: 22.19
image-classification-on-imagenetDiscreteViT#255Top 1 Accuracy: 85.07%
image-classification-on-objectnetViT-B (Discrete 512x512)#30Top-1 Accuracy: 46.62

Rank counts only results with a code link.