RandLA-Net 调通代码2：Windows 11 + Tensorflow 2.4 + Semantic KITTI数据集

一、克隆大佬的源代码

克隆 https://github.com/QingyongHu/RandLA-Net

cd RandLA-Net

二、安装 Tensorflow

找配套的python、CUDA和cudnn。TF现在是真lj。

python = 3.8

CUDA = 11.0

cudnn = 8.0

pip3 install tensorflow-gpu==2.4.4

三、重要的 lib

pip install cython==0.29.15

pip install -U scikit-learn

pip install pandas==0.25.3

四、邻近搜索和下采样 sh compile_op.sh【具体命令如下】

pip install cython==0.29.15

4.1. KNN

cd utils/nearest_neighbors

python setup.py install --home="."

cd ../../

结果：/utils/nearest_neighbors 下生成 四个文件夹和knn.cpp

4.2. subsampling

cd utils/cpp_wrappers/cpp_subsampling

python setup.py build_ext --inplace

cd ../../../

结果：/utils/cpp_wrappers/cpp_subsampling 下生成 build文件夹和grid_subsampling.cp38-win_amd64.pyd

五、准备 Semantic KITTI 数据集

5.1. 下载地址

http://semantic-kitti.org/dataset.html#download

5.2. 三个文件都下载

5.3. 解压后，所有的 dataset 文件夹合并移动到 /data/semantic_kitti/dataset[手动新建文件夹]

./dataset/sequences/00

./dataset/sequences/00/labels/*.labels

./dataset/sequences/00/velodyne/*.bin

./dataset/sequences/00/calib.txt

./dataset/sequences/00/poses.txt

./dataset/sequences/00/times.txt

pip install -U scikit-learn

pip install pandas==0.25.3

5.3. 运行 python utils/data_prepare_semantickitti.py 过程中的坑

坑：[helper_tool.py]导入错误

填坑：/utils/nearest_neighbors/lib/python 下的文件夹：KNN_NanoFLANN-0.0.0-py3.8-win-amd64.egg 名称改为 KNN_NanoFLANN

填坑：将[helper_tool.py] 中的 import nearest_neighbors.lib.python.nearest_neighbors as nearest_neighbors

改为 import utils.nearest_neighbors.lib.python.KNN_NanoFLANN.nearest_neighbors as nearest_neighbors

填坑：将[helper_tool.py] 中的