search

facebookresearch / ToMe

A method to increase the speed and lower the memory footprint of existing vision transformers.
Other
931 stars 67 forks source link

pretrained model for ViT-L on Kinetics-400 #20

Closed ChenMnZ closed 1 year ago

ChenMnZ commented 1 year ago

Thank you for this work！

Could you share the pretrained model for ViT-L on Kinetics-400?

dbolya commented 1 year ago

Hi! We actually found that training wasn't necessary to get good accuracy on the video models, so I suggest you use an existing pretrained checkpoint.