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 bysorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
WS3D_LR_ODpermissive0.355 10.833 10.754 10.611 10.067 10.699 10.036 10.149 10.425 10.177 10.248 10.033 10.355 10.095 10.143 10.750 10.245 10.600 10.167 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_LR_DET0.096 30.667 20.374 20.007 30.006 30.271 30.000 20.004 20.050 20.010 30.005 30.000 30.005 20.039 20.004 20.181 30.023 30.078 30.000 3
CSC_LR_DET0.126 20.667 20.293 30.082 20.007 20.328 20.000 20.000 30.030 30.016 20.006 20.000 20.001 30.000 30.000 30.363 20.073 20.400 20.000 4
Scratch_LR_DET0.000 40.000 40.000 40.000 40.000 40.001 40.000 20.000 30.000 40.001 40.001 40.000 30.000 40.000 40.000 30.000 40.000 40.000 40.000 2