| """ | |
| in1k_vit.py | |
| Vision Transformers trained / finetuned on ImageNet (ImageNet-21K =>> ImageNet-1K) | |
| """ | |
| from prismatic.models.backbones.vision.base_vision import TimmViTBackbone | |
| # Registry =>> Supported Vision Backbones (from TIMM) | |
| IN1K_VISION_BACKBONES = { | |
| "in1k-vit-l": "vit_large_patch16_224.augreg_in21k_ft_in1k", | |
| } | |
| class IN1KViTBackbone(TimmViTBackbone): | |
| def __init__(self, vision_backbone_id: str, image_resize_strategy: str, default_image_size: int = 224) -> None: | |
| super().__init__( | |
| vision_backbone_id, | |
| IN1K_VISION_BACKBONES[vision_backbone_id], | |
| image_resize_strategy, | |
| default_image_size=default_image_size, | |
| ) | |