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 reconstructions scenario.




Method Infoavg ap 50%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
sort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysorted bysort bysort bysort bysort by
WS3D_LR_ODpermissive0.369 10.917 10.829 10.655 10.102 10.674 10.035 20.172 10.294 10.188 10.251 10.032 10.394 10.149 10.176 10.698 10.259 10.682 20.143 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_LR_DET0.291 20.667 20.665 20.421 20.091 20.569 30.023 30.158 20.140 40.089 30.102 30.013 20.364 20.146 20.123 20.643 20.201 20.770 10.058 2
PointContrast_LR_DET0.260 30.667 20.631 30.312 30.077 30.621 20.100 10.059 30.258 20.121 20.111 20.002 30.328 30.009 40.093 30.589 30.139 30.523 30.034 3
Scratch_LR_DET0.159 40.667 20.460 40.131 40.024 40.411 40.001 40.029 40.196 30.051 40.045 40.000 40.061 40.022 30.005 40.458 40.105 40.191 40.001 4