FLPCDet: Federated Learning Point CLoud Detection

Federated Learning Point CLoud Detection (FLPCDet) is based on Carla Simulation and OpenPCDet (See README_PCDET.md).

Install FLPCDet

• FLPCDet is based on OpenPCDet (https://github.com/open-mmlab/OpenPCDet)
• OpenPCDet is based on SPCONV (https://github.com/traveller59/spconv)
• Test Environment: Ubuntu 20.04 (Python 3.8) and CUDA 11.3 (Nvidia Driver 470)

1. Install dependencies, e.g.,

pip3 install torch==1.10.0+cu113 torchvision==0.11.1+cu113 torchaudio==0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
pip install cvxpy

2. Install OpenPCDet

cd FLPCDet
pip3 install -r requirements.txt

3. Install SPCONV For CUDA 11.3, use

pip3 install spconv-cu113

4. Install OPEN3D

pip3 install open3d

5. Build

python setup.py develop

Quick Start

1. Prepare dataset

Download our example dataset first and extract to data folder: FLPCDet.tar.gz

python -m pcdet.datasets.kitti.kitti_dataset create_kitti_infos tools/cfgs/dataset_configs/tesla339_dataset.yaml

1. Federated learning under resource contraints for SECOND

python flcav_second.py -w 4096 -l 4096 --batch_size=16 --epoch=20

1. Model Folders

• Fedmodel: Cloud and edge federated models
• Output: Local models at each autonomous vehicles

1. Testing

cd tools;
python test.py --cfg_file cfgs/kitti_models/town05_test/vehicle1232.yaml --ckpt ../fedmodel/cloud/global.pth

Acknowledgement

• OpenPCDet