3D Semantic Instance with Limited Annotations Benchmark

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 semantic instance with limited annotations scenario.




Method Infoavg ap 25%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
WS3D_LA_Inspermissive0.793 11.000 10.894 50.845 40.808 10.830 20.564 10.819 10.771 20.604 20.674 10.635 10.592 31.000 10.912 10.815 40.760 11.000 10.748 1
Kangcheng Liu: WS3D: Weakly Supervised 3D Scene Segmentation with Region-Level Boundary Awareness and Instance Discrimination. European Conference on Computer Vision (ECCV), 2022
Box2Mask_LA0.755 21.000 10.943 10.860 30.694 40.872 10.505 40.681 20.943 10.695 10.666 20.431 20.600 10.857 20.580 50.875 30.741 21.000 10.648 2
Julian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll: Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes. ECCV 2022
CSC_LA_INS0.702 31.000 10.909 20.867 10.703 30.704 30.550 20.649 30.653 30.506 40.572 30.245 30.500 50.835 30.824 20.921 10.697 31.000 10.507 3
Scratch_LA_INS0.662 40.903 50.900 40.867 10.711 20.612 40.550 20.591 40.427 50.487 50.567 40.180 40.596 20.777 40.794 30.812 50.656 50.997 40.486 5
PointContrast_LA_INS0.645 51.000 10.905 30.798 50.659 50.607 50.315 50.470 50.477 40.547 30.544 50.173 50.592 30.735 50.687 40.910 20.692 40.997 40.494 4