Installation

Install the devkit of Mixed Signals dataset

git clone github.com/quan-dao/mixed-signals-devkit
cd mixed-signals-devkit
pip install -r requirements.txt
python setup.py develop

Install OpenCOOD

git clone <this_repo>
cd OpenCOOD
pip install -r requirements.txt
python opencood/utils/setup.py build_ext --inplace
python opencood/pcdet_utils/setup.py build_ext --inplace
python setup.py develop

Prepare Mixed Signals dataset

Download Mixed Signales dataset

Download our dataset from ???. After unpacking the ???.zip file, the structure of the dataset directory is as following

mixed-signals
├── Odometry
│   ├── mini_17
│   ├── mini_21
│   ├── mini_4
│   └── ...
├── PointClouds
│   ├── mini_17
│   ├── mini_21
│   ├── mini_4
│   └── ...
└── V2X_dataset-v0.4-labels.json

The Odometry directory contains csv files that store the localisation of vehicles existing in each segment of our dataset. For example,

mini_17
├── odometry_003.csv
├── odometry_004.csv
└── odometry_laser.csv

The PointClouds directory contains pcd files that store the point cloud of each agent, including 3 vehicles and 2 road-side lidars, at every syncrhonized time step of each segment. For example,

mini_17
├── top_1_1712121711.218860737.pcd
├── dome_1_1712121711.224186066.pcd
├── 003_1_1712121711.188474014.pcd
├── 004_1_1712121711.204070535.pcd
├── laser_1_1712121711.215227296.pcd
├── ...

The naming convetion of point cloud files is

<agent_name>_<index_of_synchronized_timestep_in_a_segment>_<time_stamp_this_pointcloud>.pcd

The time stamp contains two parts:

• before the decimal point: seconds
• after the decimal point: nanoseconds

Preprocess ground truths

cd OpenCOOD
python opencood/tools/msig_dataset_preprocessing.py

This script creates train_gt_infos.pkl and test_gt_infos.pkl in the root directory of Mixed Signals dataset.

Evaluate our models

Download check points from ??? and put them in OpenCOOD/ckpt

Create a directory to store the evaluation result

cd OpenCOOD
mkdir output

Evaluate Early fusion

cd OpenCOOD
python opencood/tools/inference.py --hypes_yaml opencood/hypes_yaml/msig_pillar_early.yaml --ckpt pointpillar_early_fusion.pth --fusion_method early --save_eval_result_to_directory output

Late Fusion

cd OpenCOOD
python opencood/tools/inference.py --hypes_yaml opencood/hypes_yaml/msig_pillar_late.yaml --ckpt pointpillar_no_fusion_net_epoch20.pth --fusion_method late --save_eval_result_to_directory output