S3E-SLAM

Sparse Spatial Scene Embedding powered SLAM system.

Dependencies

Here is the dependencies

for g2o:

sudo apt install libsuitesparse-dev\
                  qtdeclarative5-dev\
                  qt5-qmake\
                  libqglviewer-dev\
                  cmake\
                  libeigen3-dev

for pangolin:

sudo apt install libgl1-mesa-dev\
                  libglew-dev\
                  ffmpeg libavcodec-dev libavutil-dev libavformat-dev libswscale-dev\
pip install pyopengl

conda install -c dglteam dgl-cuda10.2

Build Instruction

Build the g2o and pangolin from source

• g2o

cd third_party
cd g2opy
mkdir build
cd build
cmake .. -DPYTHON_EXECUTABLE=$(which python)
make -j12

• pangolin

cd third_party
cd pangolin
mkdir build
cd build
cmake .. -DPYTHON_EXECUTABLE=$(which python)
make -j8
cd ..
python setup.py install 

Dataset Prepare

Please follow the following steps to build the dataset

• run the Vins-RGBD to generate the pose graph and point cloud map data
• run the dataset preprocess script to cook the dataset (crop the patch, calculate ground truth iou)

Credit

The project is based on multiple open-source projects including:

• rgbd_ptam
• vins_rgbd