lingkai/deit

Data efficient Image Transformers.

DeiT small model pretrained on imagenet with 300 epochs:

• Acc@1 79.250 Acc@5 94.604 loss 0.927 *.

DeiT small model finetuned on imagenet with 15 epochs:

• Acc@1 81.606 Acc@5 95.848 loss 0.80915 *.