Intel / article

PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing

Source: unknown
Date: Published Jan 29, 2026

Context

PaddleOCR-VL-1.5 targets documents that arrive crooked, warped or photographed from a screen. Baidu's document-parsing model adds polygon-shaped localization, allowing it to mark irregular text regions rather than assuming clean rectangular areas. The release article reports testing robustness under scanning artifacts, skew, warping, screen photography and changing illumination.

The model remains at 0.9 billion parameters and reports a 94.5% score on OmniDocBench v1.5. Alongside the original model's document recognition, the upgrade addresses a distinct practical problem: locating and reading content when the page image itself is distorted. The score describes the named test, not the share of arbitrary documents it will parse perfectly.

Terms in this piece · Glossary

benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.

Source link unavailable.unknown