Skip to content
← Models
Open weightsOpenDataLab/10 trustVerified

DocLayout-YOLO

A real-time YOLO-v10 layout detector for diverse documents, pre-trained on a synthetic 300K-page corpus with global-to-local perception.

Visit official page →

About

DocLayout-YOLO is a layout analysis model in the most literal sense: given a document page, it draws boxes around titles, paragraphs, tables, figures, and other elements, in real time, using a YOLO-v10 backbone. Two ideas carry it. First, DocSynth-300K, a large synthetic document dataset built by treating page synthesis as a two-dimensional bin-packing problem (Mesh-candidate BestFit), which diversifies pre-training far beyond scanned-paper corpora. Second, a global-to-local controllability module that keeps detection precise across wildly different element scales.

It comes from OpenDataLab, the group behind MinerU and OmniDocBench, and slots naturally into document-parsing pipelines as the 'where are things' stage before recognition. A Hugging Face demo space is live.

The license is AGPL-3.0, which is copyleft: fine for research and internal use, but anyone shipping it inside a commercial product needs to read the terms carefully. Benchmarks are author-reported (paper arXiv 2410.12628).

Specs

Context window

Image input (real-time detector; no text window)

Pricing

Free / open weights (AGPL-3.0)

Free quota

Fully open — no API

API access

No official API — self-host from Hugging Face (opendatalab/DocLayout-YOLO collection); live demo on Hugging Face Spaces

Weights

Open weights

Released

Not disclosed

Benchmarks

  • Author-reported (arXiv 2410.12628): leading results on document layout benchmarks including the D4LA family (provider-claimed)
  • Real-time inference via the YOLO-v10 backbone (provider-claimed)
  • Pre-trained on DocSynth-300K, a 300K-page synthetic document corpus
  • No independent replication published yet

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

Use cases

layout-analysisdocument-parsingpdf-processingtable-detection

Trust score — 7/10

/10 trustVerified

Scored 7/10 (TIER verified, metrics-v1). Evidence: arXiv paper 2410.12628 from OpenDataLab (the MinerU/OmniDocBench group) with released code, weights, synthetic dataset, and a live HF demo. Strengths: open and reproducible, real-time, genuinely novel synthetic pre-training approach. Docked for: (1) benchmark claims are author-reported with no independent replication; (2) AGPL-3.0 license is copyleft, which limits commercial embedding. License: AGPL-3.0.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
6.0/10
Maintenance · recency, releases, issue load
7.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · arXiv paper 2410.12628 from OpenDataLab (MinerU/OmniDocBench group); YOLO-v10 layout detector with DocSynth-300K synthetic pre-training.
  • · Released code, weights, dataset, and a live Hugging Face demo space.
  • · Benchmark claims are author-reported with no independent replication; AGPL-3.0 is copyleft.

Sources