RepSurf - Surface Representation for Point Clouds
[CVPR 2022 Oral]

By Haoxi Ran* , Jun Liu, Chengjie Wang ( * : corresponding contact)

The pytorch official implementation of "Surface Representation for Point Clouds". The repos of RepSurf on other tasks are coming:

arXiv | PDF

Other Tasks:

3D Segmentation (ongoing): RepSurf for Segmentation

Preparation

Environment

We tested under the environment:

• python 3.7
• pytorch 1.6.0
• cuda 10.1
• gcc 7.2.0
• h5py

For anaconda user, initialize the environment by:

sh init.sh

Or you can manually install the above packages and compile the cuda-based point operators by:

cd modules/pointops
python3 setup.py install

Classification

ScanObjectNN

• Performance:

• To download dataset:

wget http://download.cs.stanford.edu/orion/scanobjectnn/h5_files.zip
unzip h5_files.zip
ln -s [PATH]/h5_files data/ScanObjectNN

Note: We conduct all experiments on the hardest variant of ScanObjectNN (PB_T50_RS).

• To train Umbrella RepSurf on ScanObjectNN:

sh scripts/repsurf/scanobjectnn/repsurf_ssg_umb.sh

• To train Umbrella RepSurf (2x setting) on ScanObjectNN:

sh scripts/repsurf/scanobjectnn/repsurf_ssg_umb_2x.sh

Visualization

We provide several visualization results in the folder ./visualization for a closer look at the construction of RepSurf.

TODO

• Classification on ModelNet40
• Segmentation on S3DIS / ScanNet

Acknowledgment

We use part of the library pointops from PointWeb.

License

RepSurf is under the Apache-2.0 license. Please contact the primary author Haoxi Ran ([email protected]) for commercial use.