The 3D semantic instance prediction task involves detecting and segmenting the object in an 3D scan mesh.

Evaluation and metrics

Similarly to the ScanNet benchmark in ScanNet200 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) for all 200 categories. Note that multiple predictions of the same ground truth instance are penalized as false positives.



This table lists the benchmark results for the ScanNet200 3D semantic instance scenario.




Method Infoavg ap 50%head ap 50%common ap 50%tail ap 50%alarm clockarmchairbackpackbagballbarbasketbathroom cabinetbathroom counterbathroom stallbathroom stall doorbathroom vanitybathtubbedbenchbicyclebinblackboardblanketblindsboardbookbookshelfbottlebowlboxbroombucketbulletin boardcabinetcalendarcandlecartcase of water bottlescd caseceilingceiling lightchairclockclosetcloset doorcloset rodcloset wallclothesclothes dryercoat rackcoffee kettlecoffee makercoffee tablecolumncomputer towercontainercopiercouchcountercratecupcurtaincushiondecorationdeskdining tabledish rackdishwasherdividerdoordoorframedresserdumbbelldustpanend tablefanfile cabinetfire alarmfire extinguisherfireplacefolded chairfurnitureguitarguitar casehair dryerhandicap barhatheadphonesironing boardjacketkeyboardkeyboard pianokitchen cabinetkitchen counterladderlamplaptoplaundry basketlaundry detergentlaundry hamperledgelightlight switchluggagemachinemailboxmatmattressmicrowavemini fridgemirrormonitormousemusic standnightstandobjectoffice chairottomanovenpaperpaper bagpaper cutterpaper towel dispenserpaper towel rollpersonpianopicturepillarpillowpipeplantplateplungerposterpotted plantpower outletpower stripprinterprojectorprojector screenpurserackradiatorrailrange hoodrecycling binrefrigeratorscaleseatshelfshoeshowershower curtainshower curtain rodshower doorshower floorshower headshower wallsignsinksoap dishsoap dispensersofa chairspeakerstair railstairsstandstoolstorage binstorage containerstorage organizerstovestructurestuffed animalsuitcasetabletelephonetissue boxtoastertoaster oventoilettoilet papertoilet paper dispensertoilet paper holdertoilet seat cover dispensertoweltrash bintrash cantraytubetvtv standvacuum cleanerventwardrobewashing machinewater bottlewater coolerwater pitcherwhiteboardwindowwindowsill
sort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
TD3D Scannet200permissive0.320 20.501 20.264 20.164 20.506 30.062 20.500 10.000 10.000 10.000 10.208 10.431 20.252 31.000 10.733 30.587 20.000 20.008 20.000 30.106 10.000 20.356 10.123 40.686 10.101 20.000 10.152 20.000 10.000 20.226 10.280 30.000 20.000 10.250 10.000 10.619 20.061 30.841 10.000 10.000 20.167 10.194 10.333 20.000 20.000 10.667 20.820 10.250 30.790 41.000 10.879 20.077 10.094 30.708 10.217 20.049 20.634 10.792 10.331 40.033 50.716 20.159 20.396 20.331 40.099 20.415 10.842 10.000 20.458 10.542 10.000 10.101 20.000 10.218 10.513 20.500 20.458 20.104 20.516 10.456 10.268 40.000 10.000 10.400 10.022 10.233 20.143 20.000 10.677 10.400 10.504 50.095 30.083 50.890 20.061 20.000 10.906 10.076 20.231 10.125 20.000 20.003 20.792 30.881 10.000 20.098 30.125 40.498 50.459 20.063 10.715 10.000 20.241 40.000 10.396 20.063 10.605 10.000 10.000 20.000 10.448 50.629 30.202 20.967 10.250 20.038 10.192 10.185 20.083 41.000 11.000 10.857 20.000 20.470 20.012 10.565 30.798 10.621 10.111 10.500 11.000 10.017 20.509 10.000 10.008 11.000 10.525 20.000 10.000 10.332 30.679 10.264 20.333 20.267 11.000 10.549 10.299 50.387 20.328 30.744 40.000 10.000 20.435 51.000 10.283 40.000 10.196 10.817 10.000 10.472 10.222 30.123 40.560 20.156 2
Maksim Kolodiazhnyi, Anna Vorontsova, Anton Konushin, Danila Rukhovich: Top-Down Beats Bottom-Up in 3D Instance Segmentation. WACV 2024
Mask3D Scannet2000.388 10.542 10.357 10.237 10.610 10.091 10.125 50.000 10.000 10.000 10.065 30.668 10.451 11.000 10.955 10.640 10.500 10.039 10.125 20.063 20.409 10.311 20.291 10.609 30.266 10.000 10.163 10.000 10.008 10.044 20.496 11.000 10.000 10.018 20.000 10.756 10.573 10.808 20.000 10.010 10.042 30.130 30.552 10.042 10.000 11.000 10.725 40.750 10.883 11.000 10.832 40.024 20.107 10.614 30.226 10.250 10.628 20.792 10.677 20.400 10.741 10.278 10.511 10.077 50.111 10.313 20.715 20.302 10.017 30.200 20.000 10.188 10.000 10.178 20.736 11.000 10.615 10.514 10.409 20.380 50.600 10.000 10.000 10.400 10.013 20.254 10.381 10.000 10.123 40.400 10.839 10.258 10.463 10.926 10.265 10.000 10.857 20.099 10.021 20.500 10.027 10.028 11.000 10.502 50.016 10.076 40.500 10.612 10.578 10.005 20.597 20.194 10.497 10.000 10.500 10.000 20.323 40.000 11.000 10.000 10.748 10.708 20.050 40.890 21.000 10.008 20.151 30.301 11.000 11.000 10.792 30.945 11.000 10.511 10.004 20.753 10.776 20.287 20.020 20.003 40.974 30.033 10.412 50.000 10.000 20.000 20.667 10.000 10.000 10.491 10.676 20.352 10.335 10.060 20.822 50.527 21.000 10.517 10.606 10.853 10.000 10.004 10.806 11.000 10.727 10.000 10.042 20.739 20.000 10.399 30.391 10.504 10.591 10.571 1
Jonas Schult, Francis Engelmann, Alexander Hermans, Or Litany, Siyu Tang, Bastian Leibe: Mask3D for 3D Semantic Instance Segmentation. ICRA 2023
Minkowski 34D Inst.permissive0.203 50.369 40.134 50.078 50.479 40.003 40.500 10.000 10.000 10.000 10.100 20.371 30.300 20.667 40.746 20.400 30.000 20.000 30.000 30.031 30.000 20.074 40.165 30.413 50.000 40.000 10.070 40.000 10.000 20.000 30.221 50.000 20.000 10.000 30.000 10.372 50.070 20.706 40.000 10.000 20.000 50.123 40.033 50.000 20.000 10.422 50.732 30.000 40.778 51.000 10.845 30.000 30.090 40.636 20.000 30.000 30.158 40.000 30.250 50.050 40.693 30.123 40.051 50.385 30.009 40.118 50.406 50.000 20.000 40.200 20.000 10.000 30.000 10.133 40.307 50.500 20.251 40.000 40.281 30.402 40.317 20.000 10.000 10.000 30.000 30.060 40.000 30.000 10.396 20.200 30.669 20.021 40.218 40.720 50.000 30.000 10.696 30.025 40.000 30.000 30.000 20.000 30.125 50.596 20.000 20.191 10.500 10.595 20.369 40.000 30.500 40.000 20.143 50.000 10.000 30.000 20.226 50.000 10.000 20.000 10.701 20.511 40.000 50.851 40.000 30.000 30.150 40.052 50.100 30.981 30.500 40.286 30.000 20.000 50.000 30.545 40.522 50.250 30.000 30.000 50.522 50.000 30.500 20.000 10.000 20.000 20.282 50.000 10.000 10.178 50.382 40.018 50.056 40.000 30.997 30.107 50.677 20.313 40.000 40.726 50.000 10.000 20.583 40.903 40.200 50.000 10.000 30.333 40.000 10.442 20.083 40.109 50.387 40.000 5
C. Choy, J. Gwak, S. Savarese: 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks. CVPR 2019
CSC-Pretrain Inst.permissive0.209 40.361 50.157 40.085 40.506 20.007 30.500 10.000 10.000 10.000 10.000 50.093 50.221 40.667 40.524 50.400 30.000 20.000 30.000 30.004 40.000 20.000 50.109 50.589 40.000 40.000 10.059 50.000 10.000 20.000 30.322 20.000 20.000 10.000 30.000 10.405 30.055 40.700 50.000 10.000 20.028 40.091 50.083 30.000 20.000 10.667 20.768 20.000 40.807 31.000 10.776 50.000 30.000 50.340 50.000 30.000 30.103 50.000 30.750 10.200 30.634 50.053 50.246 30.677 20.006 50.198 30.432 40.000 20.000 40.050 40.000 10.000 30.000 10.111 50.356 40.500 20.188 50.000 40.220 40.448 20.050 50.000 10.000 10.000 30.000 30.032 50.000 30.000 10.396 20.000 40.573 40.000 50.228 30.747 40.000 30.000 10.573 50.021 50.000 30.000 30.000 20.000 30.500 40.573 30.000 20.000 50.125 40.592 30.364 50.000 30.450 50.000 20.364 20.000 10.000 30.000 20.340 30.000 10.000 20.000 10.610 30.833 10.221 10.702 50.000 30.000 30.135 50.094 40.125 20.571 40.500 40.143 50.000 20.125 30.000 30.618 20.667 40.115 50.000 30.125 21.000 10.000 30.500 20.000 10.000 20.000 20.502 40.000 10.000 10.312 40.248 50.050 40.000 50.000 30.997 30.420 30.500 40.149 50.451 20.748 20.000 10.000 20.636 30.667 50.600 20.000 10.000 30.278 50.000 10.333 40.000 50.294 20.381 50.110 3
Ji Hou, Benjamin Graham, Matthias Nie├čner, Saining Xie: Exploring Data-Efficient 3D Scene Understanding with Contrastive Scene Contexts. CVPR 2021
LGround Inst.permissive0.246 30.413 30.170 30.130 30.455 50.003 50.500 10.000 10.000 10.000 10.017 40.333 40.111 51.000 10.681 40.400 30.000 20.000 31.000 10.003 50.000 20.167 30.190 20.637 20.067 30.000 10.081 30.000 10.000 20.000 30.264 40.000 20.000 10.000 30.000 10.387 40.031 50.754 30.000 10.000 20.151 20.135 20.056 40.000 20.000 10.582 40.589 50.500 20.815 21.000 10.903 10.000 30.097 20.588 40.000 30.000 30.234 30.000 30.500 30.400 10.682 40.156 30.159 40.750 10.046 30.125 40.660 30.000 20.200 20.000 50.000 10.000 30.000 10.164 30.402 30.500 20.373 30.025 30.143 50.426 30.317 20.000 10.000 10.000 30.000 30.063 30.000 30.000 10.000 50.000 40.575 30.250 20.241 20.772 30.000 30.000 10.653 40.034 30.000 30.000 30.000 20.000 31.000 10.561 40.000 20.100 20.500 10.541 40.452 30.000 30.581 30.000 20.364 20.000 10.000 30.000 20.571 20.000 10.000 20.000 10.568 40.511 40.167 30.857 30.000 30.000 30.164 20.112 30.000 50.530 51.000 10.286 30.000 20.125 30.000 30.464 50.706 30.208 40.000 30.125 20.744 40.000 30.500 20.000 10.000 20.000 20.511 30.000 10.000 10.344 20.541 30.068 30.333 20.000 31.000 10.196 40.533 30.318 30.000 40.748 30.000 10.000 20.690 21.000 10.400 30.000 10.000 30.667 30.000 10.333 40.333 20.270 30.399 30.083 4
David Rozenberszki, Or Litany, Angela Dai: Language-Grounded Indoor 3D Semantic Segmentation in the Wild.