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 50%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
sort bysort bysort bysort bysort bysort bysort bysorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
Scratch_LA_DET0.242 40.667 20.505 30.297 30.081 20.472 40.003 40.224 10.271 40.134 40.077 40.017 20.187 40.000 40.081 20.694 30.208 20.395 40.036 4
WS3D_LA_ODpermissive0.372 10.917 10.825 10.727 10.081 10.693 10.014 30.180 20.380 10.170 30.253 10.027 10.347 10.189 30.118 10.706 20.249 10.681 20.135 2
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_DET0.306 20.667 20.720 20.426 20.059 40.560 30.053 10.138 30.301 30.191 10.130 30.015 40.220 30.326 10.053 40.727 10.149 40.642 30.127 3
PointContrast_LA_DET0.276 30.667 20.464 40.224 40.063 30.590 20.024 20.088 40.355 20.186 20.143 20.015 30.251 20.259 20.070 30.558 40.160 30.717 10.139 1