The 3D semantic instance prediction task involves detecting and segmenting the object in an 3D scan mesh.

Evaluation and metrics

Our evaluation ranks all methods according to the average precision for each class. We report the mean average precision AP at overlap 0.25 (AP 25%), overlap 0.5 (AP 50%), and over overlaps in the range [0.5:0.95:0.05] (AP). Note that multiple predictions of the same ground truth instance are penalized as false positives.



This table lists the benchmark results for the 3D object detection with limited annotations scenario.




Method Infoavg ap 50%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
sort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysorted bysort bysort bysort bysort bysort bysort by
WS3D_LA_ODpermissive0.366 11.000 10.825 10.698 10.068 10.665 10.007 40.210 10.409 10.173 10.189 10.019 10.375 10.062 40.176 10.730 10.227 10.587 20.162 1
Kangcheng Liu: WS3D: Weakly Supervised 3D Scene Segmentation with Region-Level Boundary Awareness and Instance Discrimination. European Conference on Computer Vision (ECCV), 2022
CSC_LA_DET0.274 20.667 20.458 40.189 20.034 30.528 20.093 10.098 30.329 20.117 30.147 20.004 20.318 20.154 20.136 20.687 20.166 20.733 10.072 3
PointContrast_LA_DET0.246 30.667 20.587 20.173 30.032 40.525 30.018 30.011 40.199 40.157 20.121 30.003 30.253 30.191 10.100 30.570 40.156 30.586 30.073 2
Scratch_LA_DET0.217 40.667 20.575 30.139 40.035 20.386 40.064 20.121 20.256 30.073 40.064 40.000 40.112 40.152 30.032 40.595 30.124 40.492 40.020 4