DiT

DiT is an image transformer pretrained on large-scale unlabeled document images. It learns to predict the missing visual tokens from a corrupted input image. The pretrained DiT model can be used as a backbone in other models for visual document tasks like document image classification and table detection.

You can find all the original DiT checkpoints under the Microsoft organization.

The example below demonstrates how to classify an image with Pipeline or the AutoModel class.

import torch
from transformers import pipeline

pipeline = pipeline(
    task="image-classification",
    model="microsoft/dit-base-finetuned-rvlcdip",
    torch_dtype=torch.float16,
    device=0
)
pipeline("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/dit-example.jpg")

Notes

• The pretrained DiT weights can be loaded in a [BEiT] model with a modeling head to predict visual tokens.
  Copied

  from transformers import BeitForMaskedImageModeling
  
  model = BeitForMaskedImageModeling.from_pretraining("microsoft/dit-base")

Resources

• Refer to this notebook for a document image classification inference example.