arxiv-struct-100k / README.md
TankNee's picture
Upload dataset
f08c2f7 verified
metadata
dataset_info:
  features:
    - name: id
      dtype: string
    - name: image
      dtype: string
    - name: input
      dtype: string
    - name: system
      dtype: string
    - name: output
      dtype: string
    - name: boxes
      sequence:
        sequence: float16
    - name: box_texts
      sequence: string
    - name: detail
      struct:
        - name: arxiv_id
          dtype: string
        - name: language
          dtype: string
        - name: page_idx
          dtype: int64
  splits:
    - name: train
      num_bytes: 1123448292
      num_examples: 100069
  download_size: 667611548
  dataset_size: 1123448292
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

ArXiv Paper Structure Dataset

数据集的图片内容来自arXiv的论文PDF,图片的结构标注来自SimpleTex的模型识别,识别结果已经经过一次简单的筛选,剔除部分错的太明显的样本,不过剩余样本中也有一些存在错位的样本。

英文的图片是原始的arXiv论文,中文的部分使用翻译模型进行自动翻译得到,可能存在不准确的内容。