OpenCV Zoo (C++ version)

This is the C++ version of OpenCV Zoo, which currently provides a demo and benchmark on YuNet.

Build for RISC-V arch

Xuantie C906 CPU (Allwinner D1 board)

The Allwinner D1 development board [CN] [EN] is equiped with the Xuantie C906 CPU. Xuantie C906 is a RISC-V architecture CPU with RVV intrinsic v0.7.1 support, which is developed by T-Head (平头哥半导体有限公司).

Steps to build OpenCV Zoo for Xuantie C906 CPU (Allwinner D1 board):

1. Follow this wiki on OpenCV to build OpenCV with RVV support. If you have trouble downloading toolchains, visit and download from T-Head: https://occ.t-head.cn/community/download?id=3927429448189939712 (toolchain of version 1.12.0 is recommanded).
2. Cross compile this repo using the downloaded toolchain:

   cd $opencv_zoo_cpp
   mkdir build && cd build
   cmake -D CMAKE_BUILD_TYPE=RELEASE \
         -D CMAKE_PREFIX_PATH=/path/to/opencv_riscv_install \
         -D CMAKE_TOOLCHAIN_FILE=/path/to/opencv/platforms/linux/riscv64-071-gcc.toolchain.cmake ..
   make # After this you will get two binary files: detect, benchmark

3. Download YuNet from OpenCV Zoo.
4. Send detect, benchmark, the model face_detection_yunet_2021sep.onnx and some test images to the board using adb:

   adb push detect /root/
   adb push benchmark /root/
   adb push face_detection_yunet_2021sep.onnx /root/
   adb push /path/to/image /root/

5. Run detect and benchmark on the board:

   # Shell on the board
   cd /root/
   ./detect -i=<image_name> -m=face_detection_yunet_2021sep.onnx # Run ./detect to get help
   ./benchmark -i=<image_name> -m=face_detection_yunet_2021sep.onnx # Run ./benchmark to get help