Accuracy of vit-b-16 training

google-research / big_vision

Official codebase used to develop Vision Transformer, SigLIP, MLP-Mixer, LiT and more.

Apache License 2.0

2.2k stars 147 forks source link

Accuracy of vit-b-16 training #7

Closed lucasliunju closed 2 years ago

lucasliunju commented 2 years ago

Hi, May I ask the top-1 accuracy of vit-b-16 training on imagenet-1k based on the config file "vit_1ik.py". I find the related paper report the accuracy is about 74.6.

Thank you very much!

Best Lucas

akolesnikoff commented 2 years ago

vit_i1k config with the default settings of training the B/16 model is expected to achieve around 79% top-1 accuracy. This setting, as well as other slightly better settings, are described in this paper: https://arxiv.org/abs/2106.10270.