now inside docker:

Getting Started

Data Download

Please download the nuscenes dataset in data/nuscenes with the following files present.

data
│   nuscenes
│   ├── maps
│   ├── samples
│   ├── sweeps
|   ├── v1.0-trainval

Environment Installation

We provide a dockerfile for simple setup of the environment.

# (e.g. docker build -t hsda /share/docker_files/HSDA/docker)
docker build -t $ImageName $docker_file_path/
# (e.g. docker run -it --name hsda_container --shm-size=8g --gpus all --mount type=bind,source=/share/docker_files,target=/share/code hsda /bin/bash)
docker run -it --name hsda_container --shm-size=8g --gpus all --mount type=bind,source=/share/docker_files,target=/share/code $ImageName /bin/bash
# now inside docker:
pip install --no-cache-dir -v -e .

Dataset Preparation

# Generate annotations for the nuscenes dataset.
python tools/create_data.py nuscenes --root-path data/nuscenes --out-dir data/nuscenes --extra-tag nuscenes --bev True
# Generate new dataset with HSDA shuffled camera images.
# This command may take a while.
# If it is interrupted while running, simply re-run the script and it will resume where it left off.
python prepare-hsda-dataset.py
# Generate annotations for the HSDA dataset.
python tools/create_data.py nuscenes --root-path data/nuscenes-hsda --out-dir data/nuscenes-hsda --extra-tag nuscenes --bev True

Training

# Single-GPU
python train.py $config
# Multi-GPU
./dist_train_gpu.sh $config $num
# Example: train baseline+HSDA with 2 gpus
./dist_train_gpu.sh configs/bevdet_hsda/bevdet-multi-map-aug-seg-only-6class-hsda.py 2

Testing

# We are interested only in the map results.
python test.py $config $pth --eval=bboxmap
# Example: test baseline+HSDA after training it
python test.py configs/bevdet_hsda/bevdet-multi-map-aug-seg-only-6class-hsda.py work_dirs/bevdet-multi-map-aug-seg-only-6class-hsda/epoch_20.pth --eval=bboxmap

Our pretrained pth file can be downloaded for testing here.

Bibtex

If this work is helpful for your research, please consider citing the following BibTeX entry.

@InProceedings{Glisson_2025_WACV,
    author    = {Glisson, Calvin and Chen, Qiuxiao},
    title     = {HSDA: High-Frequency Shuffle Data Augmentation for Bird's-Eye-View Map Segmentation},
    booktitle = {Proceedings of the Winter Conference on Applications of Computer Vision (WACV)},
    month     = {February},
    year      = {2025},
    pages     = {8816-8825}
}