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 50%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
sort bysort bysort bysort bysort bysort bysort bysort bysorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
Box2Mask_LA0.565 21.000 10.719 40.649 40.407 20.742 20.135 20.598 20.298 10.368 20.509 10.343 20.503 20.981 30.347 40.755 10.521 20.983 40.315 2
Julian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll: Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes. ECCV 2022
WS3D_LA_Inspermissive0.579 10.667 20.758 10.655 30.498 10.750 10.042 40.779 10.163 20.429 10.472 20.408 10.625 11.000 10.469 30.680 30.612 11.000 10.404 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_INS0.460 30.528 40.754 30.709 10.314 50.542 40.139 10.472 40.120 30.302 40.373 30.090 40.389 40.714 40.631 10.531 50.467 31.000 10.203 4
PointContrast_LA_INS0.456 40.667 20.696 50.687 20.325 40.543 30.072 30.556 30.095 40.270 50.315 40.021 50.318 51.000 10.562 20.585 40.319 50.994 30.188 5
Scratch_LA_INS0.418 50.528 40.758 20.475 50.377 30.484 50.008 50.450 50.029 50.311 30.260 50.096 30.500 30.714 40.309 50.681 20.370 40.938 50.231 3