vit#

Available models#

ViT-based models:

  • vit_small_patch16_224()

  • vit_base_patch16_224()

  • vit_large_patch16_224()

DeiT-based models:

  • deit_small_patch16_224()

  • deit_base_patch16_224()

  • deit_tiny_distilled_patch16_224()

  • deit_small_distilled_patch16_224()

  • deit_base_distilled_patch16_224()

MAE-based models:

  • vit_mae_base_patch16_224()

  • vit_mae_large_patch16_224()

DINO-based models:

  • vit_dino_small_patch8_224()

  • vit_dino_small_patch16_224()

  • vit_dino_base_patch8_224()

  • vit_dino_base_patch16_224()

Initialization of above models:

model(pretrained: bool = True, lrp: bool = False)#

Returns a ViT model.

Parameters:

  • pretrained – (bool) - If True, returns a model pre-trained on ImageNet.

  • lrp – (bool) - If True, returns a model with Layer-wise Relevance Propagation (LRP) enabled.

Example:#

>>> from txv.vit import vit_base_patch16_224
>>> model = vit_base_patch16_224(pretrained=True)
>>> model.eval()