by Long Zhuang, Taihong Yang, and Yiqing Yao
- Clone the repo:
$ git clone https://github.com/Zhuanglong2/RC-ROSNet.git- Create a conda environment using:
cd $ROOT/RCROSNet
conda env create -f env.yml
conda activate RCROSNet
pip install -e .Due to certain discrepancies with scikit library, you might need to do:
pip install scikit-image
pip install scikit-learnNOTE: We also provide requirements.txt file for venv enthusiasts.
- Dataset:
The CARRADA dataset is available on Valeo.ai's github: https://github.com/valeoai/carrada_dataset.
You must specify the path at which you store the logs and load the data from, this is done through:
cd $ROOT/RC-ROSNet-main/mvrss/utils/
python set_paths.py --carrada -dir containing the Carrada file- --logs -dir_to_output-cd $ROOT/RC-ROSNet-main/mvrss/
python -u train.py --cfg ./config_files/RC-RODNet.json --cp_store -dir_to_checkpoint_store-$ cd $ROOT/TransRadar/mvrss/
$ python -u test.py --cfg $ROOT/RC-ROSNet-main/mvrss/logs/carrada/RC-RODNet/RC-RODNet_3/config.jsonThe pre-trained weights provided were obtained using an NVIDIA GTX 3060Ti GPU. Due to the significant variance in radar data observed across different hardware platforms, we recommend that researchers perform re-evaluation on their own systems for fairness.
This repository heavily borrows from TransRadar