This is the code for paper:

Monocular Object SLAM Using Instance Segmentation with Superquadric Landmarks.

The formulas from the paper can be found in the code comments labeled as //Eq.xx.

Prerequisites

We have tested the code in Ubuntu 18.04, Pangolin 0.5, OpenCV 3.4.15, Eigen 3.3.4. This code is built based on ORB-SLAM3, so you can refer to its official environment requirements.

Building

chmod +x build.sh       
./build.sh

Examples

We provide a demo that uses the TUM rgbd_dataset_freiburg3_long_office_household sequence, please download the dataset in advance. We provide the offline instance segmentation results for this demo, which are located in ./Examples/Monocular/fr3_office_labels folder.

You can run the demo using the following command:

./Examples/Monocular/mono_tum ./Vocabulary/ORBvoc.txt ./Examples/Monocular/TUM3.yaml [path for fr3_office dataset]