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 bysorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
WS3D_LA_ODpermissive0.344 10.667 10.816 10.593 10.084 10.640 10.054 10.059 20.400 10.166 10.228 10.027 10.317 10.201 10.155 10.735 10.272 10.674 10.100 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_DET0.182 20.667 10.343 40.262 30.016 20.414 30.000 30.098 10.159 20.069 30.044 30.001 20.159 30.072 30.026 30.527 20.072 30.326 40.024 2
Scratch_LA_DET0.148 40.667 10.389 30.083 40.005 30.324 40.000 30.004 40.078 30.041 40.016 40.000 30.124 40.003 40.037 20.443 40.063 40.381 20.006 4
PointContrast_LA_DET0.162 30.167 40.503 20.273 20.004 40.446 20.000 20.030 30.077 40.078 20.049 20.000 30.231 20.113 20.013 40.501 30.075 20.346 30.016 3