Computational Efficiency Radar Object Detection based on Densely Connected Residual Block

Chih-Chung Hsu, et. al.

Official implementation of our DCSN-radar object detector. Our source is based on RODNet, thanks for the authors' contribution and great effort.

[Paper (soon)] [Dataset]

Fig.1. Training loss curves of the proposed DCSN and RODNet comparison (DCSN: Red, RODNet: Orange)

Installation

Create a conda environment for dcsn

conda create -n dcsn python=3.7 -y
conda activate dcsn

Install pytorch.

conda install pytorch torchvision -c pytorch

Install cruw-devkit package (needed for accessing the dataset) Please refer to cruw-devit repository for detailed instructions.

git clone https://github.com/yizhou-wang/cruw-devkit.git
cd cruw-devkit
pip install -e .
cd ..

Setup our DCSN package (same as RODNet does).

pip install -e .

Prepare data for our DCSN (same as RODNet does).

Note that you should set the configuration files like configs/DCSN.py to adopt DCSN architecture.

python tools/prepare_dataset/prepare_data.py \
        --config configs/<CONFIG_FILE> \
        --data_root <DATASET_ROOT> \
        --split train,test \
        --out_data_dir data/<DATA_FOLDER_NAME>

Train models

python tools/train.py --config configs/<CONFIG_FILE> \
        --data_dir data/<DATA_FOLDER_NAME> \
        --log_dir checkpoints/

Inference

python tools/test.py --config configs/<CONFIG_FILE> \
        --data_dir data/<DATA_FOLDER_NAME> \
        --checkpoint <CHECKPOINT_PATH> \
        --res_dir results/