Skip to content
← Models
Open weightsMicrosoft/10 trustVerified
Table Transformer (TATR) logo

Table Transformer (TATR)

Microsoft's DETR-based table extractor — detect tables, then rows, columns, and cells — trained on the million-table PubTables-1M corpus.

Visit official page →

About

The Table Transformer treats table extraction as object detection. Built on DETR, it comes in two flavors: one that finds tables on a page, and one that recovers the internal structure (rows, columns, cells, headers) of a cropped table. It was trained on PubTables-1M, a corpus of nearly a million fully annotated tables with canonicalized headers and quality-controlled annotations that the authors also released.

The companion GriTS metric gives table extraction a proper evaluation standard instead of the ad-hoc number matching older work used. Checkpoints for detection and structure recognition are on Hugging Face under the microsoft org, and the model is supported natively in the Transformers library, so it drops into existing pipelines with a few lines of code.

It is a 2021-era model, so newer end-to-end document parsers have passed it in breadth, but for the narrow job of table structure recognition it remains a solid, well-understood baseline.

Specs

Context window

Single page/table image input (object detection; no text window)

Pricing

Free / open weights

Free quota

Fully open — no API

API access

No official API — self-host from Hugging Face (microsoft/table-transformer-detection, microsoft/table-transformer-structure-recognition-v1.1-all)

Weights

Open weights

Released

Not disclosed

Benchmarks

  • Author-reported: state-of-the-art on PubTables-1M via the GriTS metric (paper arXiv 2110.00061, 2021)
  • Ships with the PubTables-1M dataset (575K annotated pages, 948K tables) and the GriTS evaluation metric
  • Natively supported in Hugging Face Transformers; widely used as the table-structure baseline

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

table-extractiontable-detectiondocument-parsingfinancial-documents

Trust score — 8/10

/10 trustVerified

Scored 8/10 (TIER verified, metrics-v1). Evidence: Microsoft Research paper (arXiv 2110.00061) plus the PubTables-1M dataset and GriTS metric, MIT-licensed checkpoints on Hugging Face, and native Transformers support. Strengths: fully open, the standard table-structure baseline for years, dataset plus metric plus model released together. Docked for: (1) 2021-era — newer end-to-end parsers exceed it in breadth; (2) maintenance is archival rather than active. License: MIT.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
8.0/10
Maintenance · recency, releases, issue load
6.0/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · Microsoft Research paper (arXiv 2110.00061, 2021): DETR-based table detection and structure recognition.
  • · Shipped with the PubTables-1M dataset (575K pages, 948K tables) and the GriTS evaluation metric; MIT checkpoints on HF with native Transformers support.
  • · Author-reported GriTS SOTA; 2021-era and archival, but remains the standard table-structure baseline.

Sources