Hugging Face adds PP-OCRv6 OCR model family with three tiers and up to 50-language support

PP-OCRv6 is the latest generation of PaddleOCR’s universal OCR model family, designed for text detection and recognition across documents, screenshots, multilingual images, digital displays, industrial labels, and scene text.

The model family includes three tiers—tiny (1.5M parameters), small (7.7M parameters), and medium (34.5M parameters)—each optimized for different deployment scenarios such as edge devices, mobile/desktop, and accuracy-oriented server pipelines.

The small and medium tiers support up to 50 languages, including Simplified Chinese, Traditional Chinese, English, Japanese, and 46 Latin-script languages, enabling unified multilingual OCR without separate models.

On PaddleOCR’s in-house benchmarks, PP-OCRv6_medium achieves 86.2% detection Hmean and 83.2% recognition accuracy, improving text detection by +4.6 percentage points and recognition by +5.1 percentage points over PP-OCRv5_server.

PP-OCRv6 introduces architectural and training improvements: a PPLCNetV4 unified backbone, a RepLKFPN feature pyramid network for text detection, and an EncoderWithLightSVTR recognition module that combines local context with global attention.

The release supports multiple inference backends through PaddleOCR 3.7, including a Transformers backend for Hugging Face users, ONNX Runtime for portable deployments, and native Paddle Inference, with model formats available as safetensors, Paddle inference models, and ONNX models on the Hugging Face Hub.