Submitted by kangcheng Liu.

Submission data

Full nameWS3D_LA_Ins
DescriptionData-Efficient 3D Scene Understanding
Publication titleWS3D: Weakly Supervised 3D Scene Segmentation with Region-Level Boundary Awareness and Instance Discrimination
Publication authorsKangcheng Liu
Publication venueEuropean Conference on Computer Vision (ECCV), 2022
Publication URLhttps://link.springer.com/chapter/10.1007/978-3-031-19815-1_3
Input Data TypesUses Geometry        Uses 3D
Programming language(s)C++
HardwareCore i5 CPU, A Single GTX 1080Ti GPU
Submission creation date21 Aug, 2021
Last edited2 Aug, 2023

3D semantic instance results with limited annotations



Infoavg ap 25%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
permissive0.7301.0000.9390.7030.6380.8220.4640.7560.7270.5800.6010.5450.5300.8570.8190.8560.6860.9910.623