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 bysort bysort bysorted bysort bysort bysort bysort bysort bysort bysort bysort by
Box2Mask_LA0.498 21.000 10.658 20.675 10.264 40.675 20.065 20.446 40.261 20.308 30.441 10.304 20.488 21.000 10.135 50.576 40.335 30.921 50.410 1
Julian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll: Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes. ECCV 2022
WS3D_LA_Inspermissive0.548 10.667 20.766 10.540 20.446 10.754 10.018 30.628 10.370 10.419 10.436 20.407 10.450 30.714 20.533 20.724 10.596 10.991 20.407 2
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_INS0.400 40.667 20.637 40.411 40.335 30.534 40.018 40.526 30.031 30.318 20.327 30.056 30.488 10.260 50.555 10.602 30.278 50.957 30.196 4
CSC_LA_INS0.414 30.667 20.563 50.417 30.395 20.546 30.085 10.551 20.014 40.213 40.262 40.043 40.364 40.571 30.532 30.629 20.392 20.997 10.215 3
Scratch_LA_INS0.311 50.667 20.651 30.411 50.262 50.486 50.006 50.257 50.006 50.134 50.168 50.000 50.060 50.343 40.296 40.459 50.321 40.944 40.126 5