3D Semantic Instance Results - ScanNet Benchmark

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 scenario.

Method	avg ap 50%	bathtub	bed	bookshelf	cabinet	chair	counter	curtain	desk	door	otherfurniture	picture	refrigerator	shower curtain	sink	sofa	table	toilet	window

OneFormer3D	0.801 1	1.000 1	0.973 1	0.909 4	0.698 6	0.928 2	0.582 1	0.668 26	0.685 9	0.780 2	0.687 7	0.698 9	0.702 9	1.000 1	0.794 5	0.900 2	0.784 7	0.986 42	0.635 3

UniPerception	0.800 2	1.000 1	0.930 4	0.872 8	0.727 2	0.862 14	0.454 10	0.764 13	0.820 1	0.746 4	0.706 5	0.750 1	0.772 7	0.926 34	0.764 8	0.818 17	0.826 1	0.997 32	0.660 2

Queryformer	0.787 3	1.000 1	0.933 3	0.601 38	0.754 1	0.886 7	0.558 3	0.661 28	0.767 4	0.665 9	0.716 3	0.639 15	0.808 2	1.000 1	0.844 1	0.897 3	0.804 3	1.000 1	0.624 5

MAFT	0.786 4	1.000 1	0.894 9	0.807 15	0.694 8	0.893 5	0.486 6	0.674 24	0.740 5	0.786 1	0.704 6	0.727 3	0.739 8	1.000 1	0.707 14	0.849 7	0.756 14	1.000 1	0.685 1

Mask3D	0.780 5	1.000 1	0.786 32	0.716 29	0.696 7	0.885 8	0.500 5	0.714 19	0.810 3	0.672 8	0.715 4	0.679 11	0.809 1	1.000 1	0.831 3	0.833 11	0.787 6	1.000 1	0.602 11
Jonas Schult, Francis Engelmann, Alexander Hermans, Or Litany, Siyu Tang, Bastian Leibe: Mask3D for 3D Semantic Instance Segmentation. ICRA 2023
SPFormer	0.770 6	0.903 45	0.903 6	0.806 16	0.609 20	0.886 6	0.568 2	0.815 6	0.705 8	0.711 5	0.655 9	0.652 14	0.685 14	1.000 1	0.789 6	0.809 18	0.776 10	1.000 1	0.583 16
Sun Jiahao, Qing Chunmei, Tan Junpeng, Xu Xiangmin: Superpoint Transformer for 3D Scene Instance Segmentation. AAAI 2023 [Oral]
SoftGroup++	0.769 7	1.000 1	0.803 25	0.937 1	0.684 9	0.865 11	0.213 25	0.870 2	0.664 12	0.571 15	0.758 1	0.702 7	0.807 3	1.000 1	0.653 21	0.902 1	0.792 5	1.000 1	0.626 4

SIM3D	0.766 8	1.000 1	0.948 2	0.582 44	0.599 22	0.882 9	0.510 4	0.701 21	0.632 16	0.772 3	0.685 8	0.687 10	0.782 6	1.000 1	0.833 2	0.756 28	0.798 4	1.000 1	0.622 6

SoftGroup	0.761 9	1.000 1	0.808 21	0.845 10	0.716 3	0.862 13	0.243 22	0.824 4	0.655 14	0.620 10	0.734 2	0.699 8	0.791 5	0.981 28	0.716 12	0.844 8	0.769 11	1.000 1	0.594 14
Thang Vu, Kookhoi Kim, Tung M. Luu, Xuan Thanh Nguyen, Chang D. Yoo: SoftGroup for 3D Instance Segmentaiton on Point Clouds. CVPR 2022 [Oral]
ISBNet	0.757 10	1.000 1	0.904 5	0.731 27	0.678 10	0.895 3	0.458 8	0.644 32	0.670 11	0.710 6	0.620 16	0.732 2	0.650 16	1.000 1	0.756 9	0.778 21	0.779 8	1.000 1	0.614 8
Tuan Duc Ngo, Binh-Son Hua, Khoi Nguyen: ISBNet: a 3D Point Cloud Instance Segmentation Network with Instance-aware Sampling and Box-aware Dynamic Convolution. CVPR 2023
TD3D	0.751 11	1.000 1	0.774 33	0.867 9	0.621 16	0.934 1	0.404 11	0.706 20	0.812 2	0.605 13	0.633 14	0.626 16	0.690 13	1.000 1	0.640 23	0.820 14	0.777 9	1.000 1	0.612 9
Maksim Kolodiazhnyi, Anna Vorontsova, Anton Konushin, Danila Rukhovich: Top-Down Beats Bottom-Up in 3D Instance Segmentation. WACV 2024
PBNet	0.747 12	1.000 1	0.818 17	0.837 12	0.713 4	0.844 16	0.457 9	0.647 31	0.711 7	0.614 11	0.617 17	0.657 13	0.650 16	1.000 1	0.692 15	0.822 13	0.765 13	1.000 1	0.595 13
W.Zhao, Y.Yan, C.Yang, J.Ye,X.Yang,K.Huang: Divide and Conquer: 3D Instance Segmentation With Point-Wise Binarization. ICCV 2023
GraphCut	0.732 13	1.000 1	0.788 30	0.724 28	0.642 14	0.859 15	0.248 21	0.787 11	0.618 18	0.596 14	0.653 11	0.722 5	0.583 36	1.000 1	0.766 7	0.861 4	0.825 2	1.000 1	0.504 27

IPCA-Inst	0.731 14	1.000 1	0.788 31	0.884 7	0.698 5	0.788 31	0.252 20	0.760 14	0.646 15	0.511 23	0.637 13	0.665 12	0.804 4	1.000 1	0.644 22	0.778 22	0.747 16	1.000 1	0.561 20

TopoSeg	0.725 15	1.000 1	0.806 24	0.933 2	0.668 12	0.758 35	0.272 19	0.734 18	0.630 17	0.549 19	0.654 10	0.606 17	0.697 12	0.966 31	0.612 27	0.839 9	0.754 15	1.000 1	0.573 17

DKNet	0.718 16	1.000 1	0.814 18	0.782 19	0.619 17	0.872 10	0.224 23	0.751 16	0.569 22	0.677 7	0.585 21	0.724 4	0.633 27	0.981 28	0.515 37	0.819 15	0.736 17	1.000 1	0.617 7
Yizheng Wu, Min Shi, Shuaiyuan Du, Hao Lu, Zhiguo Cao, Weicai Zhong: 3D Instances as 1D Kernels. ECCV 2022
SSEC	0.707 17	1.000 1	0.850 11	0.924 3	0.648 13	0.747 38	0.162 27	0.862 3	0.572 21	0.520 21	0.624 15	0.549 20	0.649 25	1.000 1	0.560 32	0.706 38	0.768 12	1.000 1	0.591 15

HAIS	0.699 18	1.000 1	0.849 12	0.820 13	0.675 11	0.808 25	0.279 17	0.757 15	0.465 28	0.517 22	0.596 19	0.559 19	0.600 30	1.000 1	0.654 20	0.767 24	0.676 21	0.994 38	0.560 21
Shaoyu Chen, Jiemin Fang, Qian Zhang, Wenyu Liu, Xinggang Wang: Hierarchical Aggregation for 3D Instance Segmentation. ICCV 2021
SSTNet	0.698 19	1.000 1	0.697 49	0.888 6	0.556 28	0.803 26	0.387 12	0.626 34	0.417 32	0.556 18	0.585 22	0.702 6	0.600 30	1.000 1	0.824 4	0.720 37	0.692 19	1.000 1	0.509 26
Zhihao Liang, Zhihao Li, Songcen Xu, Mingkui Tan, Kui Jia: Instance Segmentation in 3D Scenes using Semantic Superpoint Tree Networks. ICCV2021
DualGroup	0.694 20	1.000 1	0.799 27	0.811 14	0.622 15	0.817 20	0.376 13	0.805 9	0.590 20	0.487 26	0.568 25	0.525 24	0.650 16	0.835 44	0.600 28	0.829 12	0.655 23	1.000 1	0.526 23

DANCENET	0.680 21	1.000 1	0.807 22	0.733 26	0.600 21	0.768 34	0.375 14	0.543 42	0.538 23	0.610 12	0.599 18	0.498 25	0.632 29	0.981 28	0.739 11	0.856 5	0.633 29	0.882 53	0.454 36

SphereSeg	0.680 21	1.000 1	0.856 10	0.744 25	0.618 18	0.893 4	0.151 28	0.651 30	0.713 6	0.537 20	0.579 24	0.430 34	0.651 15	1.000 1	0.389 47	0.744 32	0.697 18	0.991 40	0.601 12

Box2Mask	0.677 23	1.000 1	0.847 13	0.771 21	0.509 37	0.816 21	0.277 18	0.558 41	0.482 25	0.562 17	0.640 12	0.448 30	0.700 10	1.000 1	0.666 16	0.852 6	0.578 36	0.997 32	0.488 31
Julian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll: Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes. ECCV 2022
OccuSeg+instance	0.672 24	1.000 1	0.758 41	0.682 32	0.576 26	0.842 17	0.477 7	0.504 47	0.524 24	0.567 16	0.585 23	0.451 29	0.557 38	1.000 1	0.751 10	0.797 19	0.563 39	1.000 1	0.467 35
Lei Han, Tian Zheng, Lan Xu, Lu Fang: OccuSeg: Occupancy-aware 3D Instance Segmentation. CVPR2020
Mask-Group	0.664 25	1.000 1	0.822 16	0.764 24	0.616 19	0.815 22	0.139 32	0.694 23	0.597 19	0.459 30	0.566 26	0.599 18	0.600 30	0.516 54	0.715 13	0.819 16	0.635 27	1.000 1	0.603 10
Min Zhong, Xinghao Chen, Xiaokang Chen, Gang Zeng, Yunhe Wang: MaskGroup: Hierarchical Point Grouping and Masking for 3D Instance Segmentation. ICME 2022
INS-Conv-instance	0.657 26	1.000 1	0.760 39	0.667 34	0.581 24	0.863 12	0.323 15	0.655 29	0.477 26	0.473 28	0.549 28	0.432 33	0.650 16	1.000 1	0.655 19	0.738 33	0.585 35	0.944 45	0.472 34

CSC-Pretrained	0.648 27	1.000 1	0.810 19	0.768 22	0.523 35	0.813 23	0.143 31	0.819 5	0.389 35	0.422 39	0.511 32	0.443 31	0.650 16	1.000 1	0.624 25	0.732 34	0.634 28	1.000 1	0.375 43

PE	0.645 28	1.000 1	0.773 35	0.798 18	0.538 30	0.786 32	0.088 40	0.799 10	0.350 39	0.435 37	0.547 29	0.545 21	0.646 26	0.933 33	0.562 31	0.761 27	0.556 44	0.997 32	0.501 29
Biao Zhang, Peter Wonka: Point Cloud Instance Segmentation using Probabilistic Embeddings. CVPR 2021
RPGN	0.643 29	1.000 1	0.758 40	0.582 45	0.539 29	0.826 19	0.046 44	0.765 12	0.372 37	0.436 36	0.588 20	0.539 23	0.650 16	1.000 1	0.577 29	0.750 30	0.653 25	0.997 32	0.495 30
Shichao Dong, Guosheng Lin, Tzu-Yi Hung: Learning Regional Purity for Instance Segmentation on 3D Point Clouds. ECCV 2022
Dyco3D	0.641 30	1.000 1	0.841 14	0.893 5	0.531 32	0.802 27	0.115 37	0.588 39	0.448 29	0.438 34	0.537 31	0.430 35	0.550 39	0.857 36	0.534 35	0.764 26	0.657 22	0.987 41	0.568 18
Tong He; Chunhua Shen; Anton van den Hengel: DyCo3D: Robust Instance Segmentation of 3D Point Clouds through Dynamic Convolution. CVPR2021
GICN	0.638 31	1.000 1	0.895 8	0.800 17	0.480 41	0.676 43	0.144 30	0.737 17	0.354 38	0.447 31	0.400 45	0.365 40	0.700 10	1.000 1	0.569 30	0.836 10	0.599 31	1.000 1	0.473 33

PointGroup	0.636 32	1.000 1	0.765 36	0.624 36	0.505 39	0.797 28	0.116 36	0.696 22	0.384 36	0.441 32	0.559 27	0.476 27	0.596 33	1.000 1	0.666 16	0.756 29	0.556 43	0.997 32	0.513 25
Li Jiang, Hengshuang Zhao, Shaoshuai Shi, Shu Liu, Chi-Wing Fu, Jiaya Jia: PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation. CVPR 2020 [oral]
DD-UNet+Group	0.635 33	0.667 47	0.797 29	0.714 30	0.562 27	0.774 33	0.146 29	0.810 8	0.429 31	0.476 27	0.546 30	0.399 37	0.633 27	1.000 1	0.632 24	0.722 36	0.609 30	1.000 1	0.514 24
H. Liu, R. Liu, K. Yang, J. Zhang, K. Peng, R. Stiefelhagen: HIDA: Towards Holistic Indoor Understanding for the Visually Impaired via Semantic Instance Segmentation with a Wearable Solid-State LiDAR Sensor. ICCVW 2021
DENet	0.629 34	1.000 1	0.797 28	0.608 37	0.589 23	0.627 47	0.219 24	0.882 1	0.310 41	0.402 44	0.383 47	0.396 38	0.650 16	1.000 1	0.663 18	0.543 55	0.691 20	1.000 1	0.568 19

3D-MPA	0.611 35	1.000 1	0.833 15	0.765 23	0.526 34	0.756 36	0.136 34	0.588 39	0.470 27	0.438 35	0.432 41	0.358 42	0.650 16	0.857 36	0.429 43	0.765 25	0.557 42	1.000 1	0.430 38
Francis Engelmann, Martin Bokeloh, Alireza Fathi, Bastian Leibe, Matthias Nießner: 3D-MPA: Multi Proposal Aggregation for 3D Semantic Instance Segmentation. CVPR 2020
OSIS	0.605 36	1.000 1	0.801 26	0.599 39	0.535 31	0.728 40	0.286 16	0.436 51	0.679 10	0.491 24	0.433 39	0.256 44	0.404 51	0.857 36	0.620 26	0.724 35	0.510 49	1.000 1	0.539 22

AOIA	0.601 37	1.000 1	0.761 38	0.687 31	0.485 40	0.828 18	0.008 51	0.663 27	0.405 34	0.405 43	0.425 42	0.490 26	0.596 33	0.714 47	0.553 34	0.779 20	0.597 32	0.992 39	0.424 40

PCJC	0.578 38	1.000 1	0.810 20	0.583 43	0.449 44	0.813 24	0.042 45	0.603 37	0.341 40	0.490 25	0.465 36	0.410 36	0.650 16	0.835 44	0.264 53	0.694 42	0.561 40	0.889 50	0.504 28

SSEN	0.575 39	1.000 1	0.761 37	0.473 47	0.477 42	0.795 29	0.066 41	0.529 44	0.658 13	0.460 29	0.461 37	0.380 39	0.331 53	0.859 35	0.401 46	0.692 44	0.653 24	1.000 1	0.348 45
Dongsu Zhang, Junha Chun, Sang Kyun Cha, Young Min Kim: Spatial Semantic Embedding Network: Fast 3D Instance Segmentation with Deep Metric Learning. Arxiv
RWSeg	0.567 40	0.528 57	0.708 48	0.626 35	0.580 25	0.745 39	0.063 42	0.627 33	0.240 45	0.400 45	0.497 33	0.464 28	0.515 40	1.000 1	0.475 39	0.745 31	0.571 37	1.000 1	0.429 39

NeuralBF	0.555 41	0.667 47	0.896 7	0.843 11	0.517 36	0.751 37	0.029 46	0.519 45	0.414 33	0.439 33	0.465 35	0.000 63	0.484 42	0.857 36	0.287 51	0.693 43	0.651 26	1.000 1	0.485 32
Weiwei Sun, Daniel Rebain, Renjie Liao, Vladimir Tankovich, Soroosh Yazdani, Kwang Moo Yi, Andrea Tagliasacchi: NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds. WACV 2023
MTML	0.549 42	1.000 1	0.807 23	0.588 42	0.327 49	0.647 45	0.004 53	0.815 7	0.180 48	0.418 40	0.364 49	0.182 47	0.445 45	1.000 1	0.442 42	0.688 45	0.571 38	1.000 1	0.396 41
Jean Lahoud, Bernard Ghanem, Marc Pollefeys, Martin R. Oswald: 3D Instance Segmentation via Multi-task Metric Learning. ICCV 2019 [oral]
ClickSeg_Instance	0.539 43	1.000 1	0.621 52	0.300 50	0.530 33	0.698 41	0.127 35	0.533 43	0.222 46	0.430 38	0.400 44	0.365 40	0.574 37	0.938 32	0.472 40	0.659 47	0.543 45	0.944 45	0.347 46

One_Thing_One_Click	0.529 44	0.667 47	0.718 44	0.777 20	0.399 45	0.683 42	0.000 56	0.669 25	0.138 51	0.391 46	0.374 48	0.539 22	0.360 52	0.641 51	0.556 33	0.774 23	0.593 33	0.997 32	0.251 51
Zhengzhe Liu, Xiaojuan Qi, Chi-Wing Fu: One Thing One Click: A Self-Training Approach for Weakly Supervised 3D Semantic Segmentation. CVPR 2021
Sparse R-CNN	0.515 45	1.000 1	0.538 57	0.282 51	0.468 43	0.790 30	0.173 26	0.345 53	0.429 30	0.413 42	0.484 34	0.176 48	0.595 35	0.591 52	0.522 36	0.668 46	0.476 50	0.986 43	0.327 47

Occipital-SCS	0.512 46	1.000 1	0.716 45	0.509 46	0.506 38	0.611 48	0.092 39	0.602 38	0.177 49	0.346 49	0.383 46	0.165 49	0.442 46	0.850 43	0.386 48	0.618 51	0.543 46	0.889 50	0.389 42

3D-BoNet	0.488 47	1.000 1	0.672 51	0.590 41	0.301 51	0.484 58	0.098 38	0.620 35	0.306 42	0.341 50	0.259 53	0.125 51	0.434 48	0.796 46	0.402 45	0.499 57	0.513 48	0.909 49	0.439 37
Bo Yang, Jianan Wang, Ronald Clark, Qingyong Hu, Sen Wang, Andrew Markham, Niki Trigoni: Learning Object Bounding Boxes for 3D Instance Segmentation on Point Clouds. NeurIPS 2019 Spotlight
PanopticFusion-inst	0.478 48	0.667 47	0.712 47	0.595 40	0.259 54	0.550 54	0.000 56	0.613 36	0.175 50	0.250 55	0.434 38	0.437 32	0.411 50	0.857 36	0.485 38	0.591 54	0.267 60	0.944 45	0.359 44
Gaku Narita, Takashi Seno, Tomoya Ishikawa, Yohsuke Kaji: PanopticFusion: Online Volumetric Semantic Mapping at the Level of Stuff and Things. IROS 2019 (to appear)
SPG_WSIS	0.470 49	0.667 47	0.685 50	0.677 33	0.372 47	0.562 52	0.000 56	0.482 48	0.244 44	0.316 52	0.298 50	0.052 58	0.442 47	0.857 36	0.267 52	0.702 39	0.559 41	1.000 1	0.287 49

SALoss-ResNet	0.459 50	1.000 1	0.737 43	0.159 61	0.259 53	0.587 50	0.138 33	0.475 49	0.217 47	0.416 41	0.408 43	0.128 50	0.315 54	0.714 47	0.411 44	0.536 56	0.590 34	0.873 54	0.304 48
Zhidong Liang, Ming Yang, Hao Li, Chunxiang Wang: 3D Instance Embedding Learning With a Structure-Aware Loss Function for Point Cloud Segmentation. IEEE Robotics and Automation Letters (IROS2020)
MASC	0.447 51	0.528 57	0.555 55	0.381 48	0.382 46	0.633 46	0.002 54	0.509 46	0.260 43	0.361 48	0.432 40	0.327 43	0.451 44	0.571 53	0.367 49	0.639 49	0.386 51	0.980 44	0.276 50
Chen Liu, Yasutaka Furukawa: MASC: Multi-scale Affinity with Sparse Convolution for 3D Instance Segmentation.
SegGroup_ins	0.445 52	0.667 47	0.773 34	0.185 58	0.317 50	0.656 44	0.000 56	0.407 52	0.134 52	0.381 47	0.267 52	0.217 46	0.476 43	0.714 47	0.452 41	0.629 50	0.514 47	1.000 1	0.222 54
An Tao, Yueqi Duan, Yi Wei, Jiwen Lu, Jie Zhou: SegGroup: Seg-Level Supervision for 3D Instance and Semantic Segmentation. TIP 2022
3D-SIS	0.382 53	1.000 1	0.432 60	0.245 53	0.190 55	0.577 51	0.013 50	0.263 55	0.033 58	0.320 51	0.240 54	0.075 54	0.422 49	0.857 36	0.117 58	0.699 40	0.271 59	0.883 52	0.235 53
Ji Hou, Angela Dai, Matthias Niessner: 3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans. CVPR 2019
Hier3D	0.323 54	0.667 47	0.542 56	0.264 52	0.157 58	0.550 53	0.000 56	0.205 58	0.009 60	0.270 54	0.218 55	0.075 54	0.500 41	0.688 50	0.007 64	0.698 41	0.301 56	0.459 61	0.200 55
Tan: HCFS3D: Hierarchical Coupled Feature Selection Network for 3D Semantic and Instance Segmentation.
UNet-backbone	0.319 55	0.667 47	0.715 46	0.233 54	0.189 56	0.479 59	0.008 51	0.218 56	0.067 57	0.201 57	0.173 56	0.107 52	0.123 59	0.438 55	0.150 55	0.615 52	0.355 52	0.916 48	0.093 63

R-PointNet	0.306 56	0.500 59	0.405 61	0.311 49	0.348 48	0.589 49	0.054 43	0.068 61	0.126 53	0.283 53	0.290 51	0.028 59	0.219 57	0.214 58	0.331 50	0.396 61	0.275 57	0.821 56	0.245 52

Region-18class	0.284 57	0.250 63	0.751 42	0.228 56	0.270 52	0.521 55	0.000 56	0.468 50	0.008 62	0.205 56	0.127 57	0.000 63	0.068 61	0.070 62	0.262 54	0.652 48	0.323 54	0.740 57	0.173 56

SemRegionNet-20cls	0.250 58	0.333 60	0.613 53	0.229 55	0.163 57	0.493 56	0.000 56	0.304 54	0.107 54	0.147 60	0.100 59	0.052 57	0.231 55	0.119 60	0.039 60	0.445 59	0.325 53	0.654 58	0.141 59

tmp	0.248 59	0.667 47	0.437 59	0.188 57	0.153 59	0.491 57	0.000 56	0.208 57	0.094 56	0.153 59	0.099 60	0.057 56	0.217 58	0.119 60	0.039 60	0.466 58	0.302 55	0.640 59	0.140 60

3D-BEVIS	0.248 59	0.667 47	0.566 54	0.076 62	0.035 64	0.394 62	0.027 48	0.035 63	0.098 55	0.099 62	0.030 63	0.025 60	0.098 60	0.375 57	0.126 57	0.604 53	0.181 62	0.854 55	0.171 57
Cathrin Elich, Francis Engelmann, Jonas Schult, Theodora Kontogianni, Bastian Leibe: 3D-BEVIS: Birds-Eye-View Instance Segmentation.
Sem_Recon_ins	0.227 61	0.764 46	0.486 58	0.069 63	0.098 61	0.426 61	0.017 49	0.067 62	0.015 59	0.172 58	0.100 58	0.096 53	0.054 63	0.183 59	0.135 56	0.366 62	0.260 61	0.614 60	0.168 58

ASIS	0.199 62	0.333 60	0.253 63	0.167 60	0.140 60	0.438 60	0.000 56	0.177 59	0.008 61	0.121 61	0.069 61	0.004 62	0.231 56	0.429 56	0.036 62	0.445 60	0.273 58	0.333 63	0.119 62

Sgpn_scannet	0.143 63	0.208 64	0.390 62	0.169 59	0.065 62	0.275 63	0.029 47	0.069 60	0.000 63	0.087 63	0.043 62	0.014 61	0.027 64	0.000 63	0.112 59	0.351 63	0.168 63	0.438 62	0.138 61

MaskRCNN 2d->3d Proj	0.058 64	0.333 60	0.002 64	0.000 64	0.053 63	0.002 64	0.002 55	0.021 64	0.000 63	0.045 64	0.024 64	0.238 45	0.065 62	0.000 63	0.014 63	0.107 64	0.020 64	0.110 64	0.006 64

3D Semantic Instance Benchmark