INSTALL.md

Installation instructions for Ubuntu 20.04

• Make sure CUDA and cuDNN are installed. One configurations have been tested:

  • TensorFlow-gpu 1.12.0, CUDA 9.0 and cuDNN 7.6.5
    
  
  It is recommended to create an environment using conda with the mentioned configuration.
  If you are using CUDA 11.3, then use TensorFlow-gpu 1.13.0 instead.
  

• After installing Anaconda:

      conda create -n tf-gpu cudatoolkit==9.0 tensorflow-gpu==1.12.0
      conda activate tf-gpu
  

• Install the other dependencies with conda:

  • numpy
  • scikit-learn
  • psutil
  • matplotlib (for visualization - optional)
  • mayavi (for visualization - optional)
  • PyQt5 (for visualization - optional)
  • wandb (for tracking the process - optional)

Version of the used packages

Package	cudatoolkit	cudnn	tensorflow-gpu	Python	NumPy	scikit-learn	PyQt	psutil
Version	9.0	7.6.5	1.12.0	3.6.13	1.19.2	0.24.2	5.9.2	5.9.0

• Compile the customized Tensorflow operators located in tf_custom_ops. Open a terminal in this folder, and run:

      sh compile_op.sh
  

  N.B. If you installed Tensorflow in a virtual environment, it needs to be activated when running these scripts

• Compile the C++ extension module for python located in cpp_wrappers. Open a terminal in this folder, and run:

      sh compile_wrappers.sh
  

You should now be able to train Kernel-Point Convolution models

Additional Step for Ubuntu 18.04 and 20.04 (Thank to @noahtren)

• Change the -D_GLIBCXX_USE_CXX11_ABI=0 flag for each line in tf_custom_ops/compile_op.sh to -D_GLIBCXX_USE_CXX11_ABI=1

Now you are ready to run any code in KPConv folder.