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 bysorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
WS3D_LA_Inspermissive0.548 11.000 10.690 10.476 20.406 10.756 10.031 10.733 10.215 10.351 10.415 20.319 10.541 11.000 10.477 10.576 20.557 10.941 20.377 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_LA0.465 20.667 20.591 30.773 10.331 20.682 20.029 20.409 20.122 20.284 20.432 10.253 20.466 21.000 10.127 40.806 10.280 30.821 40.291 2
Julian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll: Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes. ECCV 2022
Scratch_LA_INS0.200 50.667 20.673 20.145 50.100 50.430 40.000 30.314 50.004 50.025 50.099 50.000 50.000 50.143 40.076 50.424 40.198 40.297 50.006 5
PointContrast_LA_INS0.259 40.333 50.286 50.334 40.142 40.485 30.000 30.343 40.010 40.127 30.219 30.005 40.324 30.267 30.226 20.402 50.103 50.994 10.069 4
CSC_LA_INS0.289 30.667 20.580 40.427 30.202 30.424 50.000 30.384 30.015 30.061 40.180 40.014 30.071 40.119 50.173 30.445 30.390 20.938 30.120 3