milliMap

Project | Youtube | Paper

Pytorch implementation of our method for 2D indoor dense mapping via a low-cost, off-the-shelf mmWave radar (TI AWR1443). Our method can reconstruct a dense grid map with accuracy comparable to a lidar.

See Through Smoke: Robust Indoor Mapping with Low-cost mmWave Radar
Chris Xiaoxuan Lu, Stefano Rosa, Peijun Zhao, Bing Wang, Changhao Chen, John A. Stankovic, Niki Trigoni and Andrew Markham In MobiSys 2020.

Prerequisites

Linux or macOS
Python 2.7 or 3.6
NVIDIA GPU (12G or 24G memory) + CUDA cuDNN

Getting Started

Installation

Install PyTorch and dependencies from http://pytorch.org
Install python libraries dominate.

pip install dominate

Clone this repo:

git clone https://github.com/ChristopherLu/milliMap
cd milliMap

Dataset

To train and test a model on the full dataset, please download it from the here (dropbox link). After downloading and unzip, please put the dataset folder in this project directory.

Testing

Please download the pre-trained milliMap model from here (dropbox link), and put it under ./checkpoints/full_line_10/
For example, the following scripts tests the model on building A (bash ./scripts/test_cross_floor.sh):

python test.py --name full_line_10 --dataroot ./dataset/test/cross_floor --label_nc 0 --gpu_ids 1 --batchSize 4 --loadSize 256 --fineSize 256 --no_instance --which_epoch best

The test results will be summarized to a html file here: ./results/full_line_10/test_best/index.html, with generated images in the same level of folder.

More example scripts can be found in the scripts directory.

Training

Train a model (bash ./scripts/train_millimap.sh) in which a line detector is used and the hyper-parameter of prior loss is set to 10:

python train.py --name full_line_10 \
    --dataroot ./dataset/train/ \
    --label_nc 0 \
    --gpu_ids 0 \
    --batchSize 16 \
    --loadSize 256 \
    --fineSize 256 \
    --no_instance \
    --lambda_prior 10 \
    --detector_type line \
    --tf_log

To view training results, please checkout intermediate results in ./checkpoints/full_line_10/web/index.html.
If you have tensorflow installed, you can see tensorboard logs in ./checkpoints/full_line_10/logs by adding --tf_log to the training scripts.
If you have multiple GPUs, you can use them to speed up training by --gpu_ids 0,1,2,3,4,5,6,7 to the training scripts. Note: this is not tested and we trained our model using single GPU only. Please use at your own discretion.
You can play around different training configuration by referring to the training options file.

Training with your own dataset

When training on customized dataset is needed, please change the dataroot filed in the training scripts accordingly. Please go to pix2pixHD first to see how to prepare a dataset in valid format.

More Training/Test Details

Flags: see options/train_options.py and options/base_options.py for all the training flags; see options/test_options.py and options/base_options.py for all the test flags.

Citation

If you find this useful for your research, please use the following.

@inproceedings{lu2020millimap,
  title={See Through Smoke: Robust Indoor Mapping with Low-cost mmWave Radar},
  author={Chris Xiaoxuan Lu, Stefano Rosa, Peijun Zhao, Bing Wang, Changhao Chen, John A. Stankovic, Niki Trigoni and Andrew Markham},  
  booktitle={ACM International Conference on Mobile Systems, Applications, and Services (MobiSys)},
  year={2020}
}

Acknowledgments

This code borrows heavily from pix2pixHD.

SWX-TJ/milliMap