Complex-YOLO: Real-time 3D Object Detection on Point Clouds pytorch
This is an unofficial implementation of Complex-YOLO: Real-time 3D Object Detection on Point Clouds in pytorch. A large part of this project is based on the work here:https://github.com/marvis/pytorch-yolo2
Point Cloud Preprocessing is based on:https://github.com/skyhehe123/VoxelNet-pytorch
Download the 3D KITTI detection dataset.
Camera calibration matrices of object data set (16 MB)
Training labels of object data set (5 MB)
Velodyne point clouds (29 GB)
python3 main.py
or
python train.py
train.py 对比了测试集和训练集在训练过程中的loss变化情况, 为了观察是否过拟合.
在config.py 文件中修改文件所在的路径
todo
1.训练的结果在训练集和测试集上表现的差异很大. 现在不确定原因.
2.我想增加显示中间层特征的方案