Back to Home

Table Detection & Extraction

Research-grade table detection using Microsoft's Table Transformer (TATR) — trained on PubTables-1M. Locate tables in any PDF or image, recognise rows & columns, extract structured data, and export to CSV, JSON, or Excel. Detection runs in your browser via ONNX Runtime Web. PDFs with a text layer stay fully on-device; scanned pages and images fall back to cloud OCR, which uploads each detected row.

Microsoft Table Transformer · ONNX Runtime Web · Detection on-device
Two-stage research pipelineStage 1 — Detection: table-transformer-detection locates table bounding boxes. Stage 2 — Structure: table-transformer-structure-recognition identifies rows, columns & headers. Models (~80 MB each) download once and cache in your browser. PDFs with selectable text skip OCR — extraction is instant.

Drop a PDF, PNG, JPG, TIFF or WebP

Tables are detected even in scanned documents

Max File Size: 50MB