Skip to main content

Keras (TensorFlow v2) reimplementation of ViT models.

Project description


Keras (TensorFlow v2) reimplementation of Vision Transformer model.


pip install tfvit


Default usage (without preprocessing):

from tfvit import ViTBase32384  # + 11 other variants and input preprocessing

model = ViTBase32384()  # by default will download imagenet[21k]-pretrained weights

Custom classification (with preprocessing):

from tf_keras import layers, models
from tfvit import ViTBase32224

inputs = layers.Input(shape=(224, 224, 3), dtype='uint8')
outputs = ViTBase32224(include_top=False)(inputs)
outputs = layers.Dense(100, activation='softmax')(outputs)

model = models.Model(inputs=inputs, outputs=outputs)


Code simplification:

  • Pretrain input height and width are always equal
  • Patch height and width are always equal


  title={An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
  author={Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and  Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil},

  title={Vision Transformers Need Registers},
  author={Darcet, Timothée and Oquab, Maxime and Mairal, Julien and Bojanowski, Piotr},

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

tfvit-1.2.0.tar.gz (10.5 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page