search

google-research / vision_transformer

Apache License 2.0
10.18k stars 1.27k forks source link

How can I use imagenet21k+imagenet2012/ViT-B_16.npz to inference #307

Open break0123 opened 2 months ago

break0123 commented 2 months ago

I downloaded the model： wget https://storage.googleapis.com/vit_models/imagenet21k+imagenet2012/ViT-B_16.npz -O ViT-B_16_imagenet2012.npz And I use the following code to load the model：

model_name = 'ViT-L_16' 
model_config = models_config.MODEL_CONFIGS[model_name]
model = models.VisionTransformer(num_classes=1000, **model_config)
params = checkpoint.load(f'./test_result/{model_name}_imagenet2012.npz')

An error occurred while running： ValueError: Cannot load file containing pickled data when allow_pickle=False Please help me, how do I load a trained model?