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 annotations scenario.




Method Infoavg ap 25%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
sort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysorted bysort bysort bysort bysort by
WS3D_LA_ODpermissive0.538 11.000 10.880 10.653 10.289 10.832 10.393 10.293 10.583 10.426 10.363 10.079 10.419 10.357 10.534 10.831 10.433 10.894 10.429 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_LA_DET0.246 20.667 20.497 30.265 30.014 20.580 30.005 40.006 40.169 30.148 30.140 20.002 30.005 40.353 20.288 20.598 30.128 20.488 40.074 2
CSC_LA_DET0.239 30.444 30.405 40.269 20.013 40.595 20.029 20.024 20.150 40.178 20.094 30.029 20.089 30.296 40.220 30.624 20.076 40.707 20.066 3
Scratch_LA_DET0.206 40.333 40.500 20.183 40.014 30.504 40.021 30.018 30.315 20.100 40.046 40.002 40.090 20.348 30.112 40.376 40.109 30.613 30.029 4