search

zihangJiang / TokenLabeling

Pytorch implementation of "All Tokens Matter: Token Labeling for Training Better Vision Transformers"
Apache License 2.0
425 stars 36 forks source link

Can Token labeling reach higher than annotator model? #20

Closed ErenBalatkan closed 2 years ago

ErenBalatkan commented 3 years ago

Greetings,

Thank you for this incredible research.

I would like to know if it is possible to use Token Labeling to achieve scores higher than that of the annotator model, I believe this was the case with VOLO D5 model where it achieved higher score than NFNet, model used for annotation.

zihangJiang commented 3 years ago

Yes, you can also refer to Figure 4 (Right) in our paper for more results. The performance of the model is related to its capacity as well as the accuracy of the annotator model.