Equivariant Networks for Hierarchical Structures

Authors: Renhao Wang, Marjan Albooyeh, Siamak Ravanbakhsh

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our model on two of the largest real world point cloud segmentation benchmarks, SEMANTIC3D [19] and the Stanford Large-Scale 3D Indoor Spaces ( S3DIS) [2], as well as a dataset of virtual point cloud scenes, the VKITTI benchmark [15]. As shown in Table 1, in all cases we report new state-of-the-art.
Researcher Affiliation Academia Renhao Wang, Marjan Albooyeh University of British Columbia, {renhaow,albooyeh}@cs.ubc.ca Siamak Ravanbakhsh Mc Gill University & Mila siamak@cs.mcgill.ca
Pseudocode No The paper describes mathematical forms and operations but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes We have released a PYTORCH implementation of our models at https://github.com/rw435/wreathProdNet
Open Datasets Yes We evaluate our model on two of the largest real world point cloud segmentation benchmarks, SEMANTIC3D [19] and the Stanford Large-Scale 3D Indoor Spaces ( S3DIS) [2], as well as a dataset of virtual point cloud scenes, the VKITTI benchmark [15].
Dataset Splits No The paper mentions '15 training point clouds and 15 tests point clouds' for SEMANTIC3D, but does not explicitly state the training/validation/test splits, including a validation set, in the main body. Details are deferred to Appendix C.
Hardware Specification No The paper states 'Computational resources were provided by Mila and Compute Canada.' but does not specify any exact GPU/CPU models or detailed computer specifications used for running experiments.
Software Dependencies No The paper mentions a 'PYTORCH implementation' but does not provide specific version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes Table 3 shows 'Effect of voxelization resolution' with specific values (e.g., '2', '3', '4', ... '16' voxels per dimension). The architecture is described as 'a stack of equivariant layer Eq. (4) plus Re LU nonlinearity and residual connections'.