PaddleOCR with OpenCV for Unity

PaddleOCR With OpenCV for Unity is a Unity plugin that combines PaddleOCR with OpenCV for Unity and Unity Sentis to deliver end-to-end GPU text OCR, detection, classification, and recognition on multiple platforms with the same C# codebase. PaddleOCR With OpenCV for Unity uses PaddleOCR models under the Apache License 2.0. The execution of this asset requires OpenCV for Unity 3.0.3 or later. Requires Unity 6000.0.74f1 or later. Requires Unity Sentis (com.unity.ai.inference). Requires .NET Standard 2.1. Inference uses ONNX models converted from official PaddleOCR 3.0 (PP-OCRv5) distributions and runs them through Unity Sentis (.sentis models) via OpenCV for Unity's MultiBackendDnn integration. No network connection or cloud server access is required at runtime. All OCR inference runs entirely on-device using models bundled in StreamingAssets. Cross Platform: iOS and Android support. Win, Mac, and Linux Standalone support. WebGL support. ChromeOS support. visionOS support (beta). Support for preview in the Editor. OCR Pipeline: Text Detection - Detect text regions in images using PP-OCRv5 mobile detection models with configurable letterbox input sizes (320x320, 640x640, 1280x1280). Text Classification - Classify text line orientation (0/180) using PP-LCNet classification models. Text Recognition - Recognize text in cropped regions using PP-OCRv5 mobile recognition models and character dictionaries. Multi-language Recognition - Supports multiple recognition language packs under StreamingAssets (e.g. multi, en, latin, korean, cyrillic, arabic, devanagari, th, el, eslav, te, ta). Fully Local Inference - No internet connection or cloud API calls are needed during normal operation. Detection, classification, and recognition all run locally on the target device. Easy to Use: PaddleOCRComponent - A single MonoBehaviour that loads models, runs the full det -> cls -> rec pipeline, and exposes UnityEvent / parsed results. TextDetector, TextClassifier, TextRecognizer: Lower-level classes for custom pipelines using OpenCV's Mat and dnn APIs. Input from Texture2D, image file paths, byte arrays, and OpenCV's Mat class (via PaddleOCRComponent.Submit overloads). Switch between synchronous and asynchronous inference via PaddleOCRComponent.UseAsyncInference. Switch inference between Unity Sentis (.sentis models) and OpenCV's dnn module (.onnx models) via PaddleOCRComponent.InferenceBackend. Include Many Examples: Includes a wide variety of example usage scenarios, which consist of scene files and script codes. By running these example applications, you can learn how to develop PaddleOCR applications effectively. TutorialExample: Minimal tutorial using PaddleOCRComponent.Submit(Texture2D) and parsed results. BenchmarkExample: OCR benchmark on an assigned Texture2D with timing and visualization. MultiSource2MatHelperExample: Real-time OCR from multiple video sources via MultiSource2MatHelper. HandwritingOCRExample: Handwriting canvas with periodic and on-demand OCR.