- Add support for anaconda installation
- Added some extra compilation flags: FORCE_CUDA=1 for CUDA install and FORCE_ONLY_CPU=1 for CPU only install.
- Cubic feature sampling kernel as proposed in https://arxiv.org/pdf/2006.03761
- Comilation with older GPU architecture
- Chamfer distance introduced in https://arxiv.org/pdf/1612.00603 for dense batches
- Windows support
- Develop with python 3.7
- Fixed bug in region growing related to batching
- Ball query for partial dense data on GPU was returning only the first point. Fixed now
- Clustering algorithm for PointGroup
- Instance IoU computation on CPU and GPU
- Force no ninja for the compilation
- CPU version works for MacOS
- Fix install with pip > 19
- Random memory access on cpu radius search in the degree function
- Require pytorch implicitely and log nice message when missing
- ball query returns squared distance instead of distance
- leaner Point Cloud struct that avoids copying data
- Package would not install if pytorch is not already installed