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
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
WS3D_LA_ODpermissive0.341 11.000 10.629 10.426 10.070 10.608 10.063 10.176 10.503 10.132 10.084 10.001 10.337 10.220 10.103 10.628 10.282 10.739 10.131 1
Kangcheng Liu: WS3D: Weakly Supervised 3D Scene Segmentation with Region-Level Boundary Awareness and Instance Discrimination. European Conference on Computer Vision (ECCV), 2022
PointContrast_LA_DET0.135 20.667 20.296 30.145 20.002 30.330 30.000 40.000 30.050 30.049 20.023 20.000 30.002 40.006 40.034 30.472 20.052 20.285 40.011 2
CSC_LA_DET0.135 20.444 30.336 20.029 40.001 40.356 20.008 20.000 20.011 40.045 30.010 40.000 20.032 20.011 30.043 20.458 30.028 30.602 20.010 3
Scratch_LA_DET0.098 40.167 40.253 40.074 30.002 20.257 40.004 30.000 40.080 20.038 40.013 30.000 30.006 30.143 20.002 40.243 40.017 40.473 30.001 4