@@ -2152,8 +2152,13 @@ def _cfg(url: str = '', **kwargs) -> Dict[str, Any]:
21522152 'vit_base_patch16_reg4_gap_256.untrained' : _cfg (
21532153 input_size = (3 , 256 , 256 )),
21542154
2155- 'vit_so150m_patch16_reg4_gap_256.untrained' : _cfg (
2156- input_size = (3 , 256 , 256 )),
2155+ 'vit_so150m_patch16_reg4_gap_384.sbb_e250_in12k_ft_in1k' : _cfg (
2156+ hf_hub_id = 'timm/' ,
2157+ input_size = (3 , 384 , 384 ), crop_pct = 1.0 ),
2158+ 'vit_so150m_patch16_reg4_gap_256.sbb_e250_in12k' : _cfg (
2159+ hf_hub_id = 'timm/' ,
2160+ num_classes = 11821 ,
2161+ input_size = (3 , 256 , 256 ), crop_pct = 0.95 ),
21572162 'vit_so150m_patch16_reg4_map_256.untrained' : _cfg (
21582163 input_size = (3 , 256 , 256 )),
21592164
@@ -3482,6 +3487,17 @@ def vit_so150m_patch16_reg4_gap_256(pretrained: bool = False, **kwargs) -> Visio
34823487 return model
34833488
34843489
3490+ @register_model
3491+ def vit_so150m_patch16_reg4_gap_384 (pretrained : bool = False , ** kwargs ) -> VisionTransformer :
3492+ model_args = dict (
3493+ patch_size = 16 , embed_dim = 896 , depth = 18 , num_heads = 14 , mlp_ratio = 2.572 ,
3494+ class_token = False , reg_tokens = 4 , global_pool = 'avg' , fc_norm = False ,
3495+ )
3496+ model = _create_vision_transformer (
3497+ 'vit_so150m_patch16_reg4_gap_384' , pretrained = pretrained , ** dict (model_args , ** kwargs ))
3498+ return model
3499+
3500+
34853501@register_model
34863502def vit_intern300m_patch14_448 (pretrained : bool = False , ** kwargs ) -> VisionTransformer :
34873503 model_args = dict (
0 commit comments