Submitted by Julian Chibane.

Submission data

Full nameBox2Mask weakly supervised by bounding boxes
DescriptionBox2Mask is trained only using bounding box supervision - no per-point labels are needed.
Publication titleBox2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes
Publication authorsJulian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll
Publication venueECCV 2022
Publication URLhttp://virtualhumans.mpi-inf.mpg.de/papers/chibane22Box2Mask/Chibane_Box2Mask.pdf
Input Data TypesUses Color        Uses 3D
Programming language(s)Python, Pytorch
HardwareNVIDIA Quadro RTX 8000
Websitehttps://virtualhumans.mpi-inf.mpg.de/box2mask/
Submission creation date5 Mar, 2022
Last edited7 Jul, 2022

3D semantic instance results with limited annotations



Infoavg ap 50%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
0.4650.6670.5910.7730.3310.6820.0290.4090.1220.2840.4320.2530.4661.0000.1270.8060.2800.8210.291