search

jyaacoub / FL-ViT

Looking at optimizing the performance of Vision Transformers under a federated learning environment
5 stars 0 forks source link

Transfer pre-trained weights from ViT paper for ImageNet-21k dataset? -> https://image-net.org/ #3

Closed jyaacoub closed 1 year ago

jyaacoub commented 1 year ago

Pretrained model weights here: https://console.cloud.google.com/storage/browser/vit_models/imagenet21k;tab=objects?prefix=&forceOnObjectsSortingFiltering=false

README: https://github.com/google-research/vision_transformer/blob/main/README.md#vision-transformer

jyaacoub commented 1 year ago

To convert JAX weights to pytorch: https://github.com/lukemelas/PyTorch-Pretrained-ViT/tree/master/jax_to_pytorch