mmdetection3d代码教学

安装open3d

!pip install open3d==0.11.0

安装依赖包mmcv-full、mmdet、mmsegmentation

window安装时也用openmim

!pip install openmim
!mim install mmcv-full
!mim install mmdet
!mim install mmsegmentation
!git clone https://github.com/open-mmlab/mmdetection3d.git
%cd mmdetection3d
!pip install -e .

报错后，改变click版本
pip install click==7.1.2

或者直接安装

pip install openmim
在mmdetection3d文件夹下
mim install -e .

import mmdet3d
print(mmdet3d.__version__)

下载训练好的pointpillars模型到checkpoints文件夹

下面展示一些 内联代码片。

!mkdir checkpoints 
!wget https://download.openmmlab.com/mmdetection3d/v1.0.0_models/pointpillars/hv_pointpillars_secfpn_6x8_160e_kitti-3d-3class/hv_pointpillars_secfpn_6x8_160e_kitti-3d-3class_20220301_150306-37dc2420.pth -O checkpoints/fhv_pointpillars_secfpn_6x8_160e_kitti-3d-3class_20220301_150306-37dc2420.pth

推理计算

from mmdet3d.apis import inference_detector,init_model,show_result_meshlab
#根据colab状态设置device
device=torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#device='cuda:0'
#选择模型对应的配置文件
config='configs/pointpillars/hv_pointpillars_secfpn_6x8_160e_kitti-3d-3class.py'
#选择下载好的checkpoints
checkpoint='checkpoints/hv_pointpillars_secfpn_6x8_160e_kitti-3d-3class_20220301_150306-37dc2420.pth'
#初始化模型
model=init_model(config,checkpoint,device=device)

#找一个点云数据，这里用KITTI数据集中的一个点云文件
pcd='demo/data/kitti/kitti_000008.bin'
result,data=inference_detector(model,pcd)
#可视化检测结果
out_dir='./'
show_result_meshlab(data,result,out_dir,show=True)