• install pytorch and sparceconov.

The website for sparse convolution is here

• Download the processed dataset from here and unzip it.

You would have a folder structure like the following:

.
+-- data
+-- src
+-- readme.md
+-- data.zip

• Run the srcipt with.

cd src
./sh/ad/train00.sh

Try different script in the folder sh to reproduce the results in the paper.