CDLA: A Chinese document layout analysis (CDLA) dataset

Last update: Dec 28, 2022

Related tags

Overview

CDLA: A Chinese document layout analysis (CDLA) dataset

介绍

CDLA是一个中文文档版面分析数据集，面向中文文献类（论文）场景。包含以下10个label：

正文	标题	图片	图片标题	表格	表格标题	页眉	页脚	注释	公式
Text	Title	Figure	Figure caption	Table	Table caption	Header	Footer	Reference	Equation

共包含5000张训练集和1000张验证集，分别在train和val目录下。每张图片对应一个同名的标注文件(.json)。

样例展示：

下载链接

百度云下载：https://pan.baidu.com/s/1449mhds2ze5JLk-88yKVAA, 提取码: tp0d
Google Drive Download：https://drive.google.com/file/d/14SUsp_TG8OPdK0VthRXBcAbYzIBjSNLm/view?usp=sharing

标注格式

我们的标注工具是labelme，所以标注格式和labelme格式一致。这里说明一下比较重要的字段。

"shapes": shapes字段是一个list，里面有多个dict，每个dict代表一个标注实例。

"labels": 类别。

"points": 实例标注。因为我们的标注是Polygon形式，所以points里的坐标数量可能大于4。

"shape_type": "polygon"

"imagePath": 图片路径/名

"imageHeight": 高

"imageWidth": 宽

展示一个完整的标注样例:

{
  "version":"4.5.6",
  "flags":{},
  "shapes":[
    {
      "label":"Title",
      "points":[
        [
          553.1111111111111,
          166.59259259259258
        ],
        [
          553.1111111111111,
          198.59259259259258
        ],
        [
          686.1111111111111,
          198.59259259259258
        ],
        [
          686.1111111111111,
          166.59259259259258
        ]
      ],
      "group_id":null,
      "shape_type":"polygon",
      "flags":{}
    },
    {
      "label":"Text",
      "points":[
        [
          250.5925925925925,
          298.0740740740741
        ],
        [
          250.5925925925925,
          345.0740740740741
        ],
        [
          188.5925925925925,
          345.0740740740741
        ],
        [
          188.5925925925925,
          410.0740740740741
        ],
        [
          188.5925925925925,
          456.0740740740741
        ],
        [
          324.5925925925925,
          456.0740740740741
        ],
        [
          324.5925925925925,
          410.0740740740741
        ],
        [
          1051.5925925925926,
          410.0740740740741
        ],
        [
          1051.5925925925926,
          345.0740740740741
        ],
        [
          1052.5925925925926,
          345.0740740740741
        ],
        [
          1052.5925925925926,
          298.0740740740741
        ]
      ],
      "group_id":null,
      "shape_type":"polygon",
      "flags":{}
    },
    {
      "label":"Footer",
      "points":[
        [
          1033.7407407407406,
          1634.5185185185185
        ],
        [
          1033.7407407407406,
          1646.5185185185185
        ],
        [
          1052.7407407407406,
          1646.5185185185185
        ],
        [
          1052.7407407407406,
          1634.5185185185185
        ]
      ],
      "group_id":null,
      "shape_type":"polygon",
      "flags":{}
    }
  ],
  "imagePath":"val_0031.jpg",
  "imageData":null,
  "imageHeight":1754,
  "imageWidth":1240
}

转coco格式

执行命令:

# train
python3 labelme2coco.py CDLA_dir/train train_save_path  --labels labels.txt

# val
python3 labelme2coco.py CDLA_dir/val val_save_path  --labels labels.txt

转换结果保存在train_save_path/val_save_path目录下。

labelme2coco.py取自labelme，更多信息请参考labelme官方项目

CDLA: A Chinese document layout analysis (CDLA) dataset

Related tags

Overview

CDLA: A Chinese document layout analysis (CDLA) dataset

介绍

下载链接

标注格式

转coco格式

Owner

buptlihang

GPT-3: Language Models are Few-Shot Learners

Pipeline for fast building text classification TF-IDF + LogReg baselines.

A pytorch implementation of the ACL2019 paper "Simple and Effective Text Matching with Richer Alignment Features".

Code for CodeT5: a new code-aware pre-trained encoder-decoder model.

a chinese segment base on crf

My Implementation for the paper EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks using Tensorflow

EMNLP 2021 paper "Pre-train or Annotate? Domain Adaptation with a Constrained Budget".

Python3 to Crystal Translation using Python AST Walker

Host your own GPT-3 Discord bot

ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators

Segmenter - Transformer for Semantic Segmentation

Language Models as Zero-Shot Planners: Extracting Actionable Knowledge for Embodied Agents

Continuously update some NLP practice based on different tasks.

This code is the implementation of Text Emotion Recognition (TER) with linguistic features

An Analysis Toolkit for Natural Language Generation (Translation, Captioning, Summarization, etc.)

使用pytorch+transformers复现了SimCSE论文中的有监督训练和无监督训练方法

The RWKV Language Model

Tools for curating biomedical training data for large-scale language modeling

Build Text Rerankers with Deep Language Models

Transcribing audio files using Hugging Face's implementation of Wav2Vec2 + "chain-linking" NLP tasks to combine speech-to-text with downstream tasks like translation and summarisation.