HSDA: High-frequency Shuffle Data Augmentation for Bird’s-Eye-View Map Segmentation

& RGCN: Residual Graph Convolutional Network for Bird’s-Eye-View Semantic Segmentation

**Calvin Glisson** · **Qiuxiao Chen** California State University, San Bernardino **WACV 2025** This repo provides runnable code for RGCN and the new RGCN+HSDA method.

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. ```bash

(e.g. docker build -t hsda /share/docker_files/HSDA/docker)

docker build -t $ImageName $dockerfilepath/

(e.g. docker run -it --name hsdacontainer --shm-size=8g --gpus all --mount type=bind,source=/share/dockerfiles,target=/share/code hsda /bin/bash)

docker run -it --name hsdacontainer --shm-size=8g --gpus all --mount type=bind,source=/share/dockerfiles,target=/share/code $ImageName /bin/bash

now inside docker:

pip install --no-cache-dir -v -e . ```

Dataset Preparation

```bash

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

```bash

Single-GPU

python train.py $config

Multi-GPU

./disttraingpu.sh $config $num

Example: train baseline+HSDA with 2 gpus

./disttraingpu.sh configs/bevdet_hsda/bevdet-multi-map-aug-seg-only-6class-hsda.py 2 ```

Testing

```bash

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/bevdethsda/bevdet-multi-map-aug-seg-only-6class-hsda.py workdirs/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} }