Minkowski Engine

The MinkowskiEngine is an auto-differentiation library for sparse tensors. It supports all standard neural network layers such as convolution, pooling, unpooling, and broadcasting operations for sparse tensors. For more information, please visit the documentation page.


  • Dynamic computation graph

  • Custom kernel shapes

  • Generalized sparse convolution

  • Dilated convolution

  • Multi-GPU training

  • Multi-threaded kernel map

  • Multi-threaded compilation

  • Highly-optimized GPU kernels


You can install the MinkowskiEngine without sudo using anaconda. Using anaconda is highly recommended.


We recommend python>=3.6 for installation. If you have compilation issues, please checkout the common compilation issues page first.

1. Create a conda virtual environment and install requirements.

First, follow the anaconda documentation to install anaconda on your computer.

conda create -n py3-mink python=3.7 anaconda
conda activate py3-mink
conda install openblas numpy
conda install -c bioconda google-sparsehash
conda install pytorch torchvision -c pytorch

2. Compilation and installation

conda activate py3-mink
git clone https://github.com/StanfordVL/MinkowskiEngine.git
cd MinkowskiEngine
python setup.py install

Python virtual environment

Like the anaconda installation, make sure that you install pytorch with the the same CUDA version that nvcc uses.

sudo apt install libsparsehash-dev libopenblas-dev
# within a python3 environment
pip install torch
git clone https://github.com/StanfordVL/MinkowskiEngine.git
cd MinkowskiEngine
pip install -r requirements.txt
python setup.py install

Quick Start

To use the Minkowski Engine, you first would need to import the engine. Then, you would need to define the network. If the data you have is not quantized, you would need to voxelize or quantize the (spatial) data into a sparse tensor. Fortunately, the Minkowski Engine provides the quantization function (MinkowskiEngine.utils.sparse_quantize).

Creating a Network

import MinkowskiEngine as ME

class ExampleNetwork(ME.MinkowskiNetwork):

    def __init__(self, in_feat, out_feat, D):
        super(ExampleNetwork, self).__init__(D)
        self.conv1 = ME.MinkowskiConvolution(
        self.bn1 = ME.MinkowskiBatchNorm(64)
        self.conv2 = ME.MinkowskiConvolution(
        self.bn2 = ME.MinkowskiBatchNorm(128)
        self.pooling = ME.MinkowskiGlobalPooling(dimension=D)
        self.linear = ME.MinkowskiLinear(128, out_feat)
        self.relu = ME.MinkowskiReLU(inplace=True)

    def forward(self, x):
        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.pooling(out)
        return self.linear(out)

Forward and backward using the custom network

    # loss and network
    criterion = nn.CrossEntropyLoss()
    net = ExampleNetwork(in_feat=3, out_feat=5, D=2)

    # a data loader must return a tuple of coords, features, and labels.
    coords, feat, label = data_loader()
    input = ME.SparseTensor(feat, coords=coords)
    # Forward
    output = net(input)

    # Loss
    loss = criterion(output.F, label)

Running the Examples

After installing the package, run python examples/example.py in the package root directory. For indoor semantic segmentation. run python examples/indoor.py in the package root directory.


Discussion and Documentation

For discussion and questions, please use minkowskiengine@googlegroups.com. For API and general usage, please refer to the MinkowskiEngine documentation page for more detail.

For issues not listed on the API and feature requests, feel free to submit create an issue on the github issue page.