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 bysort bysort bysorted bysort bysort by
WS3D_LR_ODpermissive0.374 10.867 10.797 10.655 10.104 10.678 10.046 10.215 10.406 10.186 10.219 10.034 10.354 10.160 10.101 10.741 10.306 10.679 10.181 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_DET0.191 20.667 20.468 30.226 20.036 20.420 30.025 20.010 30.081 30.066 30.045 20.000 20.162 30.010 30.017 20.657 20.109 20.420 30.013 2
PointContrast_LR_DET0.187 30.667 20.523 20.109 30.027 30.435 20.005 30.013 20.199 20.070 20.035 30.000 40.183 20.033 20.003 40.497 30.078 30.488 20.005 3
Scratch_LR_DET0.076 40.667 20.099 40.015 40.005 40.190 40.000 40.000 40.033 40.007 40.001 40.000 30.000 40.010 40.004 30.094 40.014 40.237 40.000 4