Paper Overview

The paper can be found at: https://arxiv.org/abs/1711.06396

Code Execution

Multiple implementasions exist:

1. A TensorFlow 2.0.0 implementation of VoxelNet.
2. An unofficial TensorFlow version (selected here).
3. A TensorFlow-based VoxelNet system for autonomous driving.

The chosen repository is: GitHub - qianguih/voxelnet

Environment Setup on autodl Server:

OS: Ubuntu 20.04 Python: 3.7

Create Conda Environment:

conda create -n voxelnet python=3.7
conda activate voxelnet

Install Dependencies:

pip3 install tensorflow==1.14.0 -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install easydict -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install scikit-build -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install opencv-python==3.4.3.18 -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install llvmlite==0.32.1 -i https://pypi.tuna.tsinghua.edu.cn/simple
python3 -m pip install --upgrade --force pip
pip3 install matplotlib -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install shapely -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install numba -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install Cython -i https://pypi.tuna.tsinghua.edu.cn/simple

Compile Cython Extension

python3 setup.py build_ext --inplace

Error Encountered:

Cannot open include file: 'numpy/arrayobject.h': No such file or directory

Solusion: Refer to: http://t.csdnimg.cn/abFlO

Compile Evaluation Script

cd kitti_eval
g++ -o evaluate_object_3d_offline evaluate_object_3d_offline.cpp

Error:

evaluate_object_3d_offline.cpp:12:10: fatal error: boost/numeric/ublas/matrix.hpp: No such file or directory

Fix:

sudo apt-get update
sudo apt-get install libboost-all-dev

Set Execution Permissions

chmod +x launch_test.sh

Data Preparation

Folder Structure

├── data
│   └── KITTI
│       ├── training
│       │   ├── calib
│       │   ├── image_2
│       │   ├── label_2
│       │   └── velodyne
│       ├── validation
│       │   ├── calib
│       │   ├── image_2
│       │   ├── label_2
│       │   └── velodyne
│       └── data_object_calib
│           ├── testing
│           │   └── calib
│           └── training
│               └── calib

Download instructions available in: Code Reproduction: VoxelNet Tensorflow Version (3D Detection) - CSDN Blog

Udpate Configuration Paths

Modify config.py:

__C.DATA_DIR = './data/KITTI'
__C.CALIB_DIR = './data/KITTI/data_object_calib/training/calib'

Modify Test Script

Update test.py to use a pre-trained vehicle model with visualization enabled:

if __name__ == '__main__':
    parser = argparse.ArgumentParser(description='testing')
    parser.add_argument('-n', '--tag', type=str, nargs='?', default='pre_trained_car',
                        help='set log tag')
    parser.add_argument('--output-path', type=str, nargs='?',
                        default='./predictions', help='results output dir')
    parser.add_argument('-b', '--single-batch-size', type=int, nargs='?', default=2,
                        help='set batch size for each gpu')
    parser.add_argument('-v', '--vis', type=bool, nargs='?', default=True,
                        help='set the flag to True if dumping visualizations')

Execute Test

python3 test.py

Results will be saved under ./predictions/vis

Evaluate Performance

./kitti_eval/evaluate_object_3d_offline data/KITTI/validation/label_2 ./predictions

Evaluation results will be stored in the predictions folder.