PP-OCRv6 on Hugging Face: Text recognition in 50 languages with models from 1.5M to 34.5M parameters
PaddleOCR (Baidu)
PP-OCRv6, the latest generation of PaddleOCR's universal OCR model family, is now available on Hugging Face. It offers three model tiers (tiny, small, medium) ranging from 1.5M to 34.5M parameters, with the medium and small tiers supporting 50 languages. The model introduces improvements in detection and recognition, achieving up to 86.2% detection Hmean and 83.2% recognition accuracy.
PP-OCRv6 is the latest generation of PaddleOCR's universal OCR model family, designed for real-world text detection and recognition across documents, screenshots, multilingual images, digital displays, industrial labels, and scene text. It scales from 1.5M to 34.5M parameters with three tiers: tiny, small, and medium. The medium and small tiers support 50 languages including Simplified Chinese, Traditional Chinese, English, Japanese, and 46 Latin-script languages. On PaddleOCR's official benchmarks, PP-OCRv6 medium achieves 86.2% detection Hmean and 83.2% recognition accuracy, improving over PP-OCRv5 server by +4.6 and +5.1 percentage points respectively. The model uses PPLCNetV4 as a unified backbone, with RepLKFPN for detection and EncoderWithLightSVTR for recognition. It can be deployed with multiple backends: Paddle Inference, Transformers, and ONNX Runtime. Models are available on Hugging Face Hub in safetensors, Paddle inference, and ONNX formats.
- Сокращения
- OCR = Optical Character Recognition — оптическое распознавание символов
- ONNX = Open Neural Network Exchange — Open Neural Network Exchange
- RAG = Retrieval-Augmented Generation — генерация с дополнением поиском
Source: Hugging Face blog —
original