FS²D: Fully Sparse Few-Shot 3D Object Detection

This is the official implementation of FS2D

This code is built upon OpenPCDet and VoxelNeXt. You can refer to the Install and Getting Started to prepare the training environment.

• Please download the official NuScenes 3D object detection dataset and organize the downloaded files as follows:

FS2D
├── data
│   ├── nuscenes
│   │   │── v1.0-trainval 
│   │   │   │── samples
│   │   │   │── sweeps
│   │   │   │── maps
│   │   │   │── v1.0-trainval  
├── pcdet
├── tools

• Install the nuscenes-devkit with version 1.0.5 by running the following command:

pip install nuscenes-devkit==1.0.5

• Generate the data infos by running the following command (it may take several hours):

# for lidar-only setting
python -m pcdet.datasets.nuscenes.nuscenes_dataset --func create_nuscenes_infos \
    --cfg_file tools/cfgs/dataset_configs/nuscenes_dataset.yaml \
    --version v1.0-trainval

Getting Started

1. Training.

   cd FS2D/tools
   python train_meta.py --cfg_file cfgs/nuscenes_models/fs2d.yaml
   

2. Evaluation.

   cd SGF3D/tools
   python test.py --cfg_file cfgs/nuscenes_models/fs2d.yaml --ckpt checkpoint_epoch_XX.pth
   

3. Visualization.

   You can refer to 3D-Detection-Tracking-Viewer or 3d-object-vis for visualization. Both of these visualization tools are based on vedo. You can also use Open3D or mayavi for visualization.