RKNN-toolkit-1.7.1 Install pit-trapping records (installed successfully)

RKNN-toolkit-1.7.1 installation stepping pit record

Preliminary investigation

• yolov5 to rknn
• rv1126 yolov5 to rknn
• RKNN-toolkit quick start operation documentation

RKNN-toolkit-1.7.1 environment configuration (Ubuntu18.04, don’t follow if you step on the trap, you can follow the Windows configuration later)

Install Anaconda3

• Go to the official to download.
• Install:

./Anacondaxxxxxx.sh

• Configure environment variables:

# 修改 vi 配置文件解决 i 无法进入 insert 模式
cd /etc/vim
chmod a+w vimrc.tiny
打开 vimrc.tiny，修改 set compatible 为 set nocompatible
# 打开 .bashrc 文件，若无初始化则在最后一行加上，否则不加
vi ./.bashrc
export PATH=$PATH:/安装目录/anaconda3/bin
# 激活
source ./.bashrc

• Create and activate a python3.6 environment:

conda create -n rv1126 python=3.6
source activate rv1126

Configure python environment

• pip changes source .

• Official python dependencies:

  •   pip install tensorflow==1.14.0 
      pip install torch==1.5.1 torchvision==0.4.0 
      pip install mxnet==1.5.0
    

  • Install tensorflow==1.14.0:

    pip install tensorflow==1.14.0 
    

  • To install torch, go to the installation package library on the official website to download the installation package and install:

    wget https://download.pytorch.org/whl/cpu/torch-1.5.1%2Bcpu-cp36-cp36m-linux_x86_64.whl
    pip install "torch-1.5.1+cpu-cp36-cp36m-lin_amd64.whl"
    

  • To install torchvision, go to the installation package library on the official website to download the installation package and install:

    wget https://download.pytorch.org/whl/cpu/torchvision-0.4.0%2Bcpu-cp36-cp36m-manylinux1_x86_64.whl
    pip install "torchvision-0.4.0+cpu-cp36-cp36m-win_amd64.whl"
    

  • Install mxnet==1.5.0:

    pip install mxnet==1.5.0
    

• Install rknn-toolkit-1.7.1:

  •   pip install rknn_toolkit-1.7.1-cp36-cp36m-linux_x86_64.whl 
    

  • The error is reported as follows:

    • 

    • 

    • Look for the version number during installation:

    • Guess: The required lmdb-0.93 and psutil-5.6.2 cannot be found on Linux.

• verify:

  • Go to the pip official website to check the lmdb installation package : it is found that there is no Linux version of lmdb-0.93.
  • Go to the pip official website to check the psutil installation package : it is found that there is no Linux version of psutil-5.6.2.

• Guess: RNKK-toolkit-1.7.1 is not yet adapted to Ubuntu.

RKNN-toolkit-1.7.1 environment configuration (Windows)

Install Anaconda3

• Go to the official website to download and install.

• Create and activate a python3.6 environment:

  •   conda create -n rv1126 python=3.6
      conda activate rv1126
    

Configure python environment

• pip changes source .

• Official python dependencies:

  •   pip install tensorflow==1.14.0 
      pip install torch==1.5.1 torchvision==0.4.0 
      pip install mxnet==1.5.0
    

• Install tensorflow==1.14.0:

  pip install tensorflow==1.14.0 
  

• To install torch, go to the installation package library on the official website to download the installation package and install:

  pip install "torch-1.5.1+cpu-cp36-cp36m-win_amd64.whl"
  

• To install torchvision, go to the installation package library on the official website to download the installation package and install:

  pip install "torchvision-0.4.0+cpu-cp36-cp36m-win_amd64.whl"
  

• Install mxnet==1.5.0:

  pip install mxnet==1.5.0
  

• Install rknn-toolkit-1.7.1:

  pip install rpip install rknn_toolkit-1.7.1-cp36-cp36m-win_amd64.whl
  

  • Error reported:

    • Because the requests that mxnet relies on conflict with the requests that rknn-toolkit relies on.

    • After trying to replace multiple versions of mxnet: I found that mxnet-1.0.0 will not replace either numpy or requests, resulting in a conflict with the rknn-toolkit dependency package. rknn-toolkit-1.7.1 also has no dependencies and reports errors. The replacement code is as follows:

    •   pip uninstall mxnet
        pip install mxnet==1.0.0