arrow
Return

ProtoSeg: A prototype-based point cloud instance segmentation method

delete2026-02-23
delete0
PRE
AI
R
Remco Royen *
L
Leon Denis
A
Adrian Munteanu
DOI:10.1016/j.image.2026.117529delete
deleteOriginal
deleteShare
deleteSave
Abstract

Abstract

En 中文
• We propose a novel prototype-based network architecture for 3D instance segmentation. • Coefficients and prototypes are learnt jointly in parallel to obtain instance masks. • A reciprocal loss was specifically developed. • A module for multi-scale coefficient retrieval, Dilated Point Inception, is presented. • ProtoSeg outperforms SOTA for S3DIS-blocks and PartNet in both time and performance.
Keywords:
prototype-based
3D instance segmentation
point cloud
coefficient learning
multi-scale retrieval

Journal

S
signal processing: image communication
IF:
0
Papers:
71
Citations:
33

Organization

No organization information available
Cited Papers

Cited Papers

Hierarchical Aggregation for 3D Instance Segmentation
err2021-10-01
err0
errOAAI
errShaoyu Chen; Jiemin Fang; Qian Zhang; Wenyu Liu; Xinggang Wang
errShare
errSave
errShare
errSave
Mask R-CNN
err2017-10-01
err0
PREAI
errKaiming He; Georgia Gkioxari; Piotr Dollar; Ross Girshick
errShare
errSave
errShare
errSave
NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds
err2023-01-01
err0
PREAI
errSun,Weiwei; Rebain,Daniel; Liao,Renjie; Tankovich,Vladimir; Yazdani,Soroosh; Yi,Kwang Moo; Tagliasacchi,Andrea
errShare
errSave
researcher View more