PaddleOCRv4_TensorRT

Enviroment

• cuda 12.4
• cudnn 8
• tensorrt-cuda12.1-trt10.1.0.27
• opencv 4.10.0 with cuda support
• opencv_freetype (optional, If you want show result in chinese)

How to run

• modify the OpenCV_DIR, fmt_DIR, TensorRT_DIR, spglog_INCLUDE_DIRS path in CmakeLists.txt
• modify the path parameters.txt in main.cpp

  mkdir build
  cd build
  cmake ..
  make -j
  ./rec

  Result

  

In my PC => 12700 + rtx2080. The time consumption is like these with 50 times model warmup and 1000 times inference.

Thanks to

• tensorrt-cpp-api for creating a easy-to-use TensorRT C++ API Tutorial.

• PaddleOCRv2_TensorRT for creating some C++ implemention of PaddleOCR preprocess and postprocess method.

• PaddleOCR for creating awesome and practical OCR tools that help users train better models and apply them into practice.