Perspective-Invariant 3D Object Detection

Ao Liang^*,1,2,3,4  Lingdong Kong^*,1  Dongyue Lu^*,1  Youquan Liu⁵  Jian Fang⁴  Huaici Zhao⁴  Wei Tsang Ooi^1
1National University of Singapore    ²University of Chinese Academy of Sciences   
³Key Laboratory of Opto-Electronic Information Processing, Chinese Academy of Sciences   
⁴Shenyang Institute of Automation, Chinese Academy of Sciences ⁵Fudan University
^*Equally contributed to this work   

Updates

• [July 2025]: Project page released.
• [June 2025]: Pi3DET has been extended to Track 5: Cross-Platform 3D Object Detection of the RoboSense Challenge at IROS 2025. See the track homepage, GitHub repo for more details.

Todo

Since the Pi3DET dataset is being used for Track 5: Cross-Platform 3D Object Detection of the RoboSense Challenge at IROS 2025, in the interest of fairness we are temporarily not releasing all of the data and annotations. If you’re interested, we have open‑sourced a subset of the data and code—please refer to the track details for more information.

• Release Phase 1 dataset of the IROS Track, which is KITTI-like single-framee format.
• Release Phase 2 dataset of the IROS Track, which is KITTI-like single-framee format.
• Release all data of Pi3DET, which has temporal information.

Download

The Track 5 dataset follows the KITTI format. Each sample consists of:

• A front-view RGB image
• A LiDAR point cloud covering the camera’s field of view
• Calibration parameters
• 3D bounding-box annotations (for training)

Calibration and annotations are packaged together in .pkl files.

We use the same training set (vehicle platform) for both phases, but different validation sets. The full dataset is hosted on Hugging Face:

robosense/track5-cross-platform-3d-object-detection

1. Download the dataset

   python tools/load_dataset.py $USER_DEFINE_OUTPUT_PATH
   

2. Link data into the project

    # Create target directory
    mkdir -p data/pi3det
   
    # Link the training split
    ln -s $USER_DEFINE_OUTPUT_PATH/track5-cross-platform-3d-object-detection/phase12_vehicle_training/training \
        data/pi3det/training
   
    # Link the validation split for Phase 1 (Drone)
    ln -s $USER_DEFINE_OUTPUT_PATH/track5-cross-platform-3d-object-detection/phase1_drone_validation/validation \
        data/pi3det/validation
   
    # Link the .pkl info files
    ln -s $USER_DEFINE_OUTPUT_PATH/track5-cross-platform-3d-object-detection/phase12_vehicle_training/training/pi3det_infos_train.pkl \
        data/pi3det/pi3det_infos_train.pkl
    ln -s $USER_DEFINE_OUTPUT_PATH/track5-cross-platform-3d-object-detection/phase1_drone_validation/validation/pi3det_infos_val.pkl \
        data/pi3det/pi3det_infos_val.pkl
   

3. Verify your directory structure
   After linking, your data/ folder should look like this:

    data/
    └── pi3det/
        ├── training/
        │   ├── image/
        │   │   ├── 0000000.jpg
        │   │   └── 0000001.jpg
        │   └── point_cloud/
        │       ├── 0000000.bin
        │       └── 0000001.bin
        ├── validation/
        │   ├── image/
        │   │   ├── 0000000.jpg
        │   │   └── 0000001.jpg
        │   └── point_cloud/
        │       ├── 0000000.bin
        │       └── 0000001.bin
        ├── pi3det_infos_train.pkl
        └── pi3det_infos_val.pkl
   

Pi3DET Dataset

Detailed statistic information

Examples

Examples