image tokenizer 原理和代码

image tokenizer原理步骤：

（1）图像分块：将输入图像划分为N×N（如16×16）的patch
（2）线性投影：通过卷积或全连接层将每个patch展平为embedding
（3）添加位置编码：将position encoding加到patch embedding

huggingface/transformers调用代码：

from transformers import ViTImageProcessor, ViTModel

processor = ViTImageProcessor.from_pretrained("google/vit-base-patch16-224")
model = ViTModel.from_pretrained("google/vit-base-patch16-224")

inputs = processor(images=image, return_tensors="pt")

参考源码：

github.com/google-research/vision_transformer

github.com/pytorch/vision/blob/main/torchvision/models/vision_transformer.py

github.com/huggingface/transformers/blob/main/src/transformers/models/vit/modeling_vit.py