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


Method Infoavg ioubathtubbedbookshelfcabinetchaircountercurtaindeskdoorfloorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwallwindow
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
Mink340.679 10.811 10.734 20.739 20.641 10.804 10.413 10.759 20.696 10.545 10.938 20.518 10.141 70.623 10.757 10.680 10.723 20.684 10.896 10.821 10.651 1
joint point-based0.621 20.645 30.746 10.612 40.571 30.795 20.386 20.798 10.485 30.539 20.943 10.445 20.287 10.520 30.418 30.635 20.744 10.570 20.859 20.795 20.628 2
DVVNet0.562 30.648 20.700 30.770 10.586 20.687 40.333 40.650 30.514 20.475 30.906 80.359 40.223 20.340 60.442 20.422 60.668 30.501 30.708 40.779 30.534 4
3DMV, FTSDF0.501 40.558 50.608 60.424 100.478 40.690 30.246 80.586 40.468 40.450 40.911 60.394 30.160 50.438 40.212 70.432 50.541 60.475 40.742 30.727 40.477 5
3DMV0.484 50.484 70.538 70.643 30.424 50.606 90.310 50.574 50.433 50.378 50.796 100.301 50.214 30.537 20.208 80.472 40.507 90.413 70.693 50.602 90.539 3
Angela Dai, Matthias Niessner: 3DMV: Joint 3D-Multi-View Prediction for 3D Semantic Scene Segmentation. ECCV'18
SurfaceConvPF0.442 60.505 60.622 50.380 110.342 80.654 60.227 100.397 70.367 70.276 70.924 40.240 70.198 40.359 50.262 50.366 70.581 40.435 50.640 60.668 70.398 6
Hao Pan, Shilin Liu, Yang Liu, Xin Tong: Convolutional Neural Networks on 3D Surfaces Using Parallel Frames.
Tangent Convolutionspermissive0.438 70.437 90.646 40.474 70.369 60.645 70.353 30.258 90.282 100.279 60.918 50.298 60.147 60.283 70.294 40.487 30.562 50.427 60.619 70.633 80.352 8
Maxim Tatarchenko, Jaesik Park, Vladlen Koltun, Qian-Yi Zhou: Tangent convolutions for dense prediction in 3d. CVPR 2018
SPLAT Netcopyleft0.393 80.472 80.511 80.606 50.311 90.656 50.245 90.405 60.328 90.197 100.927 30.227 90.000 120.001 120.249 60.271 120.510 70.383 90.593 80.699 50.267 10
Hang Su, Varun Jampani, Deqing Sun, Subhransu Maji, Evangelos Kalogerakis, Ming-Hsuan Yang, Jan Kautz: SPLATNet: Sparse Lattice Networks for Point Cloud Processing. CVPR 2018
ScanNet+FTSDF0.383 90.297 110.491 90.432 90.358 70.612 80.274 60.116 110.411 60.265 80.904 90.229 80.079 100.250 80.185 90.320 100.510 70.385 80.548 90.597 100.394 7
PointNet++permissive0.339 100.584 40.478 100.458 80.256 110.360 120.250 70.247 100.278 110.261 90.677 120.183 100.117 80.212 100.145 110.364 80.346 120.232 120.548 90.523 110.252 11
Charles R. Qi, Li Yi, Hao Su, Leonidas J. Guibas: pointnet++: deep hierarchical feature learning on point sets in a metric space.
SSC-UNetpermissive0.308 110.353 100.290 120.278 120.166 120.553 100.169 120.286 80.147 120.148 120.908 70.182 110.064 110.023 110.018 120.354 90.363 100.345 100.546 110.685 60.278 9
ScanNetpermissive0.306 120.203 120.366 110.501 60.311 90.524 110.211 110.002 120.342 80.189 110.786 110.145 120.102 90.245 90.152 100.318 110.348 110.300 110.460 120.437 120.182 12
Angela Dai, Angel X. Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, Matthias Nießner: ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes. CVPR'17

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




Method Infoavg ap 25%bathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
Sgpn_scannet0.317 10.444 30.580 20.382 10.256 10.474 20.166 20.131 10.341 10.257 10.129 20.105 20.181 30.086 30.266 20.653 10.392 10.600 10.266 1
MTML0.243 20.667 20.653 10.289 20.103 30.491 10.037 30.060 30.134 20.095 30.040 30.013 30.235 10.143 10.173 30.529 20.146 20.558 20.002 3
MaskRCNN 2d->3d Proj0.227 30.850 10.074 30.002 30.191 20.150 30.221 10.103 20.073 30.131 20.147 10.387 10.197 20.143 10.532 10.356 30.117 30.380 30.030 2

This table lists the benchmark results for the 2D semantic label scenario.


Method Infoavg ioubathtubbedbookshelfcabinetchaircountercurtaindeskdoorfloorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwallwindow
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
3DMV (2d proj)0.498 10.481 20.612 10.579 20.456 20.343 10.384 10.623 10.525 10.381 20.845 10.254 20.264 20.557 10.182 20.581 20.598 10.429 10.760 20.661 30.446 3
Angela Dai, Matthias Niessner: 3DMV: Joint 3D-Multi-View Prediction for 3D Semantic Scene Segmentation. ECCV'18
ILC-PSPNet0.475 20.490 10.581 20.289 40.507 10.067 40.379 20.610 20.417 30.435 10.822 30.278 10.267 10.503 20.228 10.616 10.533 20.375 20.820 10.729 10.560 1
Enet (reimpl)0.376 30.264 40.452 40.452 30.365 30.181 20.143 40.456 30.409 40.346 30.769 40.164 30.218 30.359 30.123 40.403 40.381 40.313 40.571 30.685 20.472 2
Re-implementation of Adam Paszke, Abhishek Chaurasia, Sangpil Kim, Eugenio Culurciello: ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation.
ScanNet (2d proj)permissive0.330 40.293 30.521 30.657 10.361 40.161 30.250 30.004 40.440 20.183 40.836 20.125 40.060 40.319 40.132 30.417 30.412 30.344 30.541 40.427 40.109 4
Angela Dai, Angel X. Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, Matthias Nießner: ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes. CVPR'17

This table lists the benchmark results for the 2D semantic instance scenario.




Method Infoavg apbathtubbedbookshelfcabinetchaircountercurtaindeskdoorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwindow
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
MaskRCNN_ScanNetpermissive0.119 10.129 10.212 10.002 10.112 10.148 10.014 10.205 10.044 10.066 10.078 10.095 10.142 10.030 10.128 10.139 10.080 10.459 10.057 1
Re-implementation of Kaiming He, Georgia Gkioxari, Piotr Dollár, Ross Girshick: Mask R-CNN. ICCV'17

This table lists the benchmark results for the scene type classification scenario.




Method Infoavg recallapartmentbathroombedroom / hotelbookstore / libraryconference roomcopy/mail roomhallwaykitchenlaundry roomliving room / loungemiscofficestorage / basement / garage
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
resnet50_scannet0.353 10.250 10.812 10.529 10.500 10.500 10.000 10.500 10.571 10.000 10.556 10.000 10.375 10.000 1