It's a unofficial tensorflow Implementation of Contfuse: Deep Continuous Fusion for Multi-Sensor 3D Object Detection. It uses C++ \ CUDA C \ Python to complete this project.
I split KITTI train data to testing data \ training data \ verification data.
kitti dataset: <-- 7481 train data
|-- data_object_calib <-- 7481
|--calib
|-- image_2 <-- 7481
|-- lidar_files <-- 7481
|-- testing
|-- label_files <-- 1000
|-- training
|-- label_files <-- 6431
|-- val
|-- label_files <-- 50
tensorflow 1.14 numpy 1.16 opencv 3.4 easydict cudnn 7.6.0 cuda 10.1 python 3.7 tqdm
step. 1 Generate dataset index document, you need modified dataset path.
cd src/scripts
python gen_dataset_idx.py
step. 2 train
python train.py
python predict.py
CONTFUSE: Deep Continuous Fusion for Multi-Sensor 3D Object Detection
PIXOR: Real-time 3D Object Detection from Point Clouds