Vision Transformer (ViT) base model, pretrained on ImageNet-21k and fine-tuned for 224x224 image classification.
google/vit-base-patch16-224 is a Vision Transformer (ViT) model for image classification. It was pretrained on ImageNet-21k (14 million images, 21,843 classes) and then fine-tuned on ImageNet 2012 (1 million images, 1,000 classes), both at 224x224 resolution. The model treats an image as a sequence of 16x16 patches, linearly embeds them, and processes the sequence with a transformer encoder.
Introduced in the paper “An Image is Worth 16x16 Words,” the weights here were converted from the timm repository. You can use it directly for classification or leverage its representations for downstream tasks with a linear layer.
The model is trained and evaluated at 224x224 resolution.
It was pretrained on ImageNet-21k (14 million images, 21,843 classes).
It was fine-tuned on ImageNet 2012 (1 million images, 1,000 classes).
The model was introduced in the paper “An Image is Worth 16x16 Words” and released by Google; the model card was written by the Hugging Face team.
