Submitted anonymously.

Submission data

Full nameSoftGroup++: Scalable 3D Instance Segmentation with Octree Pyramid Grouping
DescriptionSoftGroup++ is built upon SoftGroup, which differs in three important aspects: (1) performs octree $k$-NN instead of vanilla $k$-NN to reduce time complexity from $\mathcal{O}(n^2)$ to $\mathcal{O}(n \log n)$, (2) performs pyramid scaling that adaptively downsample backbone outputs to reduce search space for $k$-NN and grouping modules, and (3) performs late devoxelization that delay the conversion from voxel to points towards the end of the model such that intermediate components runs at low computational cost. Extensive experiments on various indoor and outdoor datasets demonstrate the efficacy of the proposed SoftGroup++. Notably, SoftGroup++ processes large scenes of millions points by a single forward without dividing the input into multiple parts, thus providing training-inference consistency. Especially, SoftGroup++ achieves 2.4 points AP$_{50}$ improvement while nearly $6\times$ faster than existing fastest method on S3DIS dataset.
Input Data TypesUses Color,Uses Geometry        Uses 3D
Programming language(s)Python, Cuda
HardwareRTX 8000
Submission creation date27 Apr, 2022
Last edited19 May, 2022

3D semantic instance results

Infoavg ap 50%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow