This is the PyTorch implementation of our paper:
When2com: Multi-Agent Perception via Communication Graph Grouping
Yen-Cheng Liu, Junjiao Tian, Nathaniel Glaser, Zsolt Kira
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020


  • Python 3.6
  • Pytorch 0.4.1
  • Other required packages in requirement.txt

Getting started

Download and install miniconda

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

Create conda environment

conda create -n semseg python=3.6
source actviate semseg

Install the required packages

pip install -r requirements.txt

Download AirSim-MAP dataset and unzip it.

  • Download the zip file you would like to run

Alt text

Move the datasets to the dataset path

mkdir dataset
mv (dataset folder name) dataset/


# [Single-request multi-support] All norm  
python train.py --config configs/srms-allnorm.yml --gpu=0

# [Multi-request multi-support] when2com model  
python train.py --config configs/mrms-when2com.yml --gpu=0


# [Single-request multi-support] All norm  
python test.py --config configs/srms-allnorm.yml --model_path <your trained weights> --gpu=0

# [Multi-request multi-support] when2com model  
python test.py --config configs/mrms-when2com.yml --model_path <your trained weights> --gpu=0


  • This work was supported by ONR grant N00014-18-1-2829.
  • This code is built upon the implementation from Pytorch-semseg.


If you find this repository useful, please cite our paper:

    title={When2com: Multi-Agent Perception via Communication Graph Grouping},
    author={Yen-Cheng Liu and Junjiao Tian and Nathaniel Glaser and Zsolt Kira},
    booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},