Skip to content

Table Extraction

OmniDocs provides multiple specialized extractors for table recognition in PDFs and images. Each extractor is implemented as a standalone class with a consistent interface, making it easy to swap between methods for different use cases.

Available Table Extractors

  • CamelotExtractor: Accurate table extraction from PDFs (lattice and stream modes).
  • PDFPlumberExtractor: Lightweight, fast PDF table extraction.
  • SuryaTableExtractor: Deep learning-based table structure recognition.
  • TabulaExtractor: Java-based PDF table extraction (requires Java runtime).
  • TableTransformerExtractor: Transformer-based table detection and extraction.
  • TableFormerExtractor: Advanced table structure parsing with deep learning.

Tutorials

Basic Usage Example

from omnidocs.tasks.table_extraction.extractors.camelot import CamelotExtractor

# Initialize extractor
extractor = CamelotExtractor()

# Extract tables from PDF
results = extractor.extract("path/to/file.pdf")

# Access tables
for i, table in enumerate(results.tables):
    print(f"Table {i+1} as DataFrame:\n", table.df)

Batch Processing Multiple PDFs

import os
from omnidocs.tasks.table_extraction.extractors.pdfplumber import PDFPlumberExtractor

extractor = PDFPlumberExtractor()
pdf_folder = "path/to/pdf_folder/"

for filename in os.listdir(pdf_folder):
    if filename.lower().endswith(".pdf"):
        pdf_path = os.path.join(pdf_folder, filename)
        try:
            results = extractor.extract(pdf_path)
            print(f"{filename}: {len(results.tables)} tables found")
        except Exception as e:
            print(f"Error processing {filename}: {e}")

Working with Different Extractors

from omnidocs.tasks.table_extraction.extractors import (
    CamelotExtractor, PDFPlumberExtractor, SuryaTableExtractor, TabulaExtractor
)

extractors = [CamelotExtractor(), PDFPlumberExtractor(), SuryaTableExtractor(), TabulaExtractor()]
pdf_path = "sample.pdf"

for extractor in extractors:
    try:
        results = extractor.extract(pdf_path)
        print(f"{extractor.__class__.__name__}: {len(results.tables)} tables")
    except Exception as e:
        print(f"{extractor.__class__.__name__} error: {e}")

Output Format

All extractors return a TableOutput object with:

class TableOutput:
    tables: List[TableResult]  # Each table as a DataFrame and metadata
    source_file: str           # Path to the processed file

Installation Requirements

Each extractor may require specific dependencies:

CamelotExtractor

pip install camelot-py[cv] pandas

PDFPlumberExtractor

pip install pdfplumber pandas

SuryaTableExtractor

pip install surya-table torch pandas

TabulaExtractor

pip install tabula-py pandas
# Requires Java installed and in PATH

TableTransformerExtractor

pip install table-transformer torch pandas

TableFormerExtractor

pip install tableformer torch pandas

Troubleshooting

1. Java Not Found (TabulaExtractor): - Ensure Java is installed and added to your system PATH.

2. No Tables Detected: - Try a different extractor or adjust parameters (e.g., lattice/stream mode for Camelot). - Check PDF quality and ensure tables are not scanned images (use OCR if needed).

3. Import Errors: - Install missing dependencies as shown above.

4. Output Not as Expected: - Inspect the DataFrame output and adjust extraction settings.

Best Practices

  1. Choose the Right Extractor:
  2. Use Camelot for vector PDFs with clear table lines.
  3. PDFPlumber works well for lightweight, fast extraction.
  4. SuryaTable, TableTransformer, or TableFormer handle complex or scanned tables effectively.

  5. Tabula is ideal when you need Java-based extraction.

  6. Optimize Performance:

  7. Batch process files and initialize extractors once.

  8. Use GPU-enabled extractors for large-scale jobs.

  9. Handle Errors Gracefully:

  10. Wrap extraction in try-except blocks.

  11. Log or print errors for debugging.

  12. Validate Output:

  13. Always inspect the DataFrame output for correctness.
  14. Post-process tables as needed for your workflow.