PP-OCRV3

Key Features:

• Curved Text Detection (EAST+DB Hybrid): Accurately reads text in circular/irregular layouts.

• Transformer-Based Recognition (SVTR): 30% better accuracy on low-quality images.

• Handwriting Support: Limited to neat handwritten text.

• Dynamic Super-Resolution: Enhances blurry/low-res inputs pre-processing.

• Enterprise-Ready: Docker containers and REST APIs available.

Model Deployment Status:

General Availability Yes (Enterprise & Open-source)

Supported Data Types for Input Image, Scanned PDF, Curved Text

Supported Data Types for Output Text + Layout Markup (JSON)

Supported # Tokens for Input Batch processing (10 images)

Supported # Tokens for Output 4k (structured text)

Knowledge Cutoff March 2023

Tool Use Limited API integration

Best For:

• Invoice/receipt digitization

• Street sign/vehicle plate recognition

• Availability:

  • PaddlePaddle Enterprise

  • Azure AI