Extracting tabular data from PDFs with Camelot & Excalibur
summary
Camelot and Excalibur claim to turn messy PDF pages into usable tables by combining PDFMiner text boxes with Ghostscript + OpenCV image processing; they ship a web UI (Excalibur) and a Python API that returns pandas DataFrames and CSVs. Two parsing modes lattice (lines via OpenCV) and stream (text alignment) cover most born‑digital PDFs, visual tools show visual debugging overlays, and accuracy is measured by textbox‑to‑cell alignment. Scanned docs need OCR (planned). The system handles hundreds of varied layouts with manual table areas as a fallback.
AI-written summary · about
chapters (9)
- 0:00Introduction and context
- 1:47History of PDF format
- 4:40Challenges of table extraction
- 9:39Introduction to Camelot
- 11:41Demonstrating Camelot
- 15:52Technical implementation
- 18:25Excalibur web interface
- 21:22Roadmap and conclusion
- 22:56Questions and answers
related talks
29mConquering PDFs- document understanding beyond plain text
31mParsing PDF: As easy as working with JSON data with PDF Extract API
06mUse Your Data Skills for Good
38mIs It A Bug Or Is It A Story?
40mPreparing data for LLM training with Data Prep Kit
49mThe CV: A Data Scientist's View
27mTrallie: Shaping Unstructured Data Into Valuable Information
26mText to Data: Make Your Code Malleable, Not Brittle