DocBank Document Layout Detection

Document layout analysis benchmark (Li et al., COLING 2020). Detect and classify document elements including titles, abstracts, paragraphs, figures, tables, and captions.

📝

text annotation

Configuration Fileconfig.yaml

# DocBank Document Layout Detection Configuration
# Based on Li et al., COLING 2020

annotation_task_name: "DocBank Document Layout Analysis"

data_files:
  - "sample-data.json"

item_properties:
  id_key: "id"
  text_key: "image_url"
  context_key: "context"

user_config:
  allow_all_users: true

annotation_schemes:
  - annotation_type: "multiselect"
    name: "document_elements"
    description: "Select all document elements visible"
    labels:
      - name: "abstract"
        tooltip: "Abstract section"
      - name: "author"
        tooltip: "Author names"
      - name: "caption"
        tooltip: "Figure/table captions"
      - name: "date"
        tooltip: "Publication date"
      - name: "equation"
        tooltip: "Mathematical equations"
      - name: "figure"
        tooltip: "Figures and images"
      - name: "footer"
        tooltip: "Page footer"
      - name: "header"
        tooltip: "Page header"
      - name: "list"
        tooltip: "Bulleted or numbered lists"
      - name: "paragraph"
        tooltip: "Body paragraphs"
      - name: "reference"
        tooltip: "Bibliography/references"
      - name: "section"
        tooltip: "Section headings"
      - name: "table"
        tooltip: "Tables"
      - name: "title"
        tooltip: "Document title"

  - annotation_type: "radio"
    name: "document_type"
    description: "What type of document is this?"
    labels:
      - name: "academic_paper"
        tooltip: "Academic/research paper"
      - name: "technical_report"
        tooltip: "Technical report"
      - name: "book_chapter"
        tooltip: "Book or chapter"
      - name: "news_article"
        tooltip: "News article"
      - name: "form"
        tooltip: "Form or application"
      - name: "invoice"
        tooltip: "Invoice or receipt"
      - name: "other"
        tooltip: "Other document type"

  - annotation_type: "radio"
    name: "layout_complexity"
    description: "How complex is the layout?"
    labels:
      - name: "single_column"
        tooltip: "Single column layout"
      - name: "double_column"
        tooltip: "Two column layout"
      - name: "multi_column"
        tooltip: "Three or more columns"
      - name: "mixed"
        tooltip: "Mixed column layouts"

  - annotation_type: "radio"
    name: "scan_quality"
    description: "Quality of the document image"
    labels:
      - name: "high"
        tooltip: "Clear, high-resolution"
      - name: "medium"
        tooltip: "Readable but some issues"
      - name: "low"
        tooltip: "Poor quality, hard to read"

  - annotation_type: "text"
    name: "element_boxes"
    description: "List bounding boxes: element_type,x1,y1,x2,y2 (one per line)"

interface_config:
  item_display_format: "<img src='{{text}}' style='max-width:100%; max-height:600px;'/><br/><small>{{context}}</small>"

output_annotation_format: "json"
output_annotation_dir: "annotations"

Sample Datasample-data.json

[
  {
    "id": "doc_001",
    "image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/8/8f/Example_academic_paper.pdf/page1-424px-Example_academic_paper.pdf.jpg",
    "context": "Document page image. Identify all layout elements: title, abstract, sections, figures, tables, captions, etc."
  },
  {
    "id": "doc_002",
    "image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/5/5e/Schloss_Neuschwanstein_2013.jpg/1200px-Schloss_Neuschwanstein_2013.jpg",
    "context": "Analyze document layout structure. Mark headers, paragraphs, lists, and other elements."
  }
]

// ... and 1 more items

Get This Design

View on GitHub

Clone or download from the repository

Quick start:

git clone https://github.com/davidjurgens/potato-showcase.git
cd potato-showcase/docbank
potato start config.yaml

Details

Annotation Types

bboxmultiselect

Domain

Document AILayout Analysis

Use Cases

Document UnderstandingLayout DetectionInformation Extraction

Related Designs

BDD100K Autonomous Driving Segmentation

Large-scale diverse driving video dataset (Yu et al., CVPR 2020). Annotate driving scenes with bounding boxes, lane markings, drivable areas, and full-frame instance segmentation.

bboxpolygon

CelebA Face Attributes Classification

Large-scale face attributes dataset with 40 binary attributes (Liu et al., ICCV 2015). Annotate celebrity face images with attributes including hair color, age, gender, and facial features.

multiselectbbox

CUB-200-2011 Fine-Grained Bird Classification

Fine-grained visual categorization of 200 bird species (Wah et al., 2011). Annotate bird images with species labels, part locations, and attribute annotations.

radiomultiselect