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 25%	bathtub	bed	bookshelf	cabinet	chair	counter	curtain	desk	door	otherfurniture	picture	refrigerator	shower curtain	sink	sofa	table	toilet	window

UniPerception	0.884 4	1.000 1	0.979 18	0.872 16	0.869 3	0.892 25	0.806 9	0.890 5	0.835 28	0.892 7	0.755 12	0.811 1	0.779 10	0.955 45	0.951 6	0.876 22	0.914 1	0.997 37	0.840 4

InsSSM	0.883 5	1.000 1	0.996 4	0.800 37	0.865 4	0.960 4	0.808 8	0.852 14	0.940 6	0.899 6	0.785 4	0.810 2	0.700 19	1.000 1	0.912 17	0.851 40	0.895 2	0.997 37	0.827 6

MG-Former	0.887 3	1.000 1	0.991 12	0.837 24	0.801 21	0.935 16	0.887 2	0.857 9	0.946 4	0.891 8	0.748 15	0.805 3	0.739 15	1.000 1	0.993 2	0.809 55	0.876 14	1.000 1	0.842 3

OneFormer3D	0.896 2	1.000 1	1.000 1	0.913 6	0.858 6	0.951 7	0.786 12	0.837 17	0.916 12	0.908 2	0.778 7	0.803 4	0.750 13	1.000 1	0.976 5	0.926 4	0.882 7	0.995 45	0.849 1
Maxim Kolodiazhnyi, Anna Vorontsova, Anton Konushin, Danila Rukhovich: OneFormer3D: One Transformer for Unified Point Cloud Segmentation.
SIM3D	0.878 8	1.000 1	0.972 23	0.863 18	0.817 18	0.952 6	0.821 6	0.783 28	0.890 17	0.902 5	0.735 18	0.797 5	0.799 9	1.000 1	0.931 14	0.893 13	0.853 20	1.000 1	0.792 15

MAFT	0.860 17	1.000 1	0.990 14	0.810 34	0.829 13	0.949 9	0.809 7	0.688 45	0.836 27	0.904 3	0.751 14	0.796 6	0.741 14	1.000 1	0.864 37	0.848 42	0.837 22	1.000 1	0.828 5

Spherical Mask(CtoF)	0.875 10	1.000 1	0.991 13	0.873 14	0.850 9	0.946 10	0.691 23	0.752 33	0.926 7	0.889 10	0.759 10	0.794 7	0.820 2	1.000 1	0.912 17	0.900 9	0.878 11	1.000 1	0.769 20

EV3D	0.877 9	1.000 1	0.996 6	0.873 14	0.854 8	0.950 8	0.691 23	0.783 29	0.926 7	0.889 11	0.754 13	0.794 8	0.820 2	1.000 1	0.912 17	0.900 9	0.860 18	1.000 1	0.779 18

Competitor-SPFormer	0.881 6	1.000 1	1.000 1	0.845 22	0.854 7	0.962 2	0.714 19	0.857 10	0.904 14	0.902 4	0.782 6	0.789 9	0.662 25	1.000 1	0.988 3	0.874 25	0.886 6	0.997 37	0.847 2

Mask3D	0.870 14	1.000 1	0.985 15	0.782 44	0.818 17	0.938 14	0.760 14	0.749 34	0.923 9	0.877 13	0.760 9	0.785 10	0.820 2	1.000 1	0.912 17	0.864 34	0.878 11	0.983 51	0.825 7
Jonas Schult, Francis Engelmann, Alexander Hermans, Or Litany, Siyu Tang, Bastian Leibe: Mask3D for 3D Semantic Instance Segmentation. ICRA 2023
Queryformer	0.874 12	1.000 1	0.978 20	0.809 35	0.876 2	0.936 15	0.702 20	0.716 39	0.920 11	0.875 14	0.766 8	0.772 11	0.818 6	1.000 1	0.995 1	0.916 6	0.892 4	1.000 1	0.767 21

TST3D	0.879 7	1.000 1	0.994 7	0.921 5	0.807 20	0.939 13	0.771 13	0.887 6	0.923 10	0.862 15	0.722 20	0.768 12	0.756 12	1.000 1	0.910 27	0.904 7	0.836 24	0.999 36	0.824 8
Duc Tran Dang Trung, Byeongkeun Kang, Yeejin Lee: MSTA3D: Multi-scale Twin-attention for 3D Instance Segmentation. ACM Multimedia 2024
ExtMask3D	0.867 15	1.000 1	1.000 1	0.756 51	0.816 19	0.940 12	0.795 10	0.760 32	0.862 22	0.888 12	0.739 16	0.763 13	0.774 11	1.000 1	0.929 15	0.878 21	0.879 9	1.000 1	0.819 11

DKNet	0.815 27	1.000 1	0.930 38	0.844 23	0.765 33	0.915 19	0.534 42	0.805 25	0.805 34	0.807 19	0.654 31	0.763 14	0.650 28	1.000 1	0.794 50	0.881 18	0.766 30	1.000 1	0.758 23
Yizheng Wu, Min Shi, Shuaiyuan Du, Hao Lu, Zhiguo Cao, Weicai Zhong: 3D Instances as 1D Kernels. ECCV 2022
PointRel	0.901 1	1.000 1	0.978 21	0.928 3	0.879 1	0.962 3	0.882 3	0.749 34	0.947 3	0.912 1	0.802 3	0.753 15	0.820 2	1.000 1	0.984 4	0.919 5	0.894 3	1.000 1	0.815 12
: Relation3D (PointRel): Enhancing Relation Modeling for Point Cloud Instance Segmentation.
GraphCut	0.832 23	1.000 1	0.922 46	0.724 55	0.798 22	0.902 23	0.701 21	0.856 12	0.859 23	0.715 27	0.706 23	0.748 16	0.640 39	1.000 1	0.934 13	0.862 35	0.880 8	1.000 1	0.729 27

SoftGroup++	0.874 12	1.000 1	0.972 24	0.947 1	0.839 12	0.898 24	0.556 38	0.913 2	0.881 20	0.756 22	0.828 2	0.748 17	0.821 1	1.000 1	0.937 11	0.937 1	0.887 5	1.000 1	0.821 9

ISBNet	0.835 21	1.000 1	0.950 32	0.731 53	0.819 15	0.918 18	0.790 11	0.740 36	0.851 26	0.831 16	0.661 30	0.742 18	0.650 28	1.000 1	0.937 10	0.814 54	0.836 23	1.000 1	0.765 22
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
SoftGroup	0.865 16	1.000 1	0.969 25	0.860 19	0.860 5	0.913 20	0.558 35	0.899 3	0.911 13	0.760 21	0.828 1	0.736 19	0.802 8	0.981 42	0.919 16	0.875 23	0.877 13	1.000 1	0.820 10
Thang Vu, Kookhoi Kim, Tung M. Luu, Xuan Thanh Nguyen, Chang D. Yoo: SoftGroup for 3D Instance Segmentaiton on Point Clouds. CVPR 2022 [Oral]
SPFormer	0.851 18	1.000 1	0.994 8	0.806 36	0.774 29	0.942 11	0.637 27	0.849 15	0.859 24	0.889 9	0.720 21	0.730 20	0.665 24	1.000 1	0.911 24	0.868 32	0.873 16	1.000 1	0.796 14
Sun Jiahao, Qing Chunmei, Tan Junpeng, Xu Xiangmin: Superpoint Transformer for 3D Scene Instance Segmentation. AAAI 2023 [Oral]
SSTNet	0.789 33	1.000 1	0.840 60	0.888 10	0.717 47	0.835 42	0.717 18	0.684 46	0.627 56	0.724 25	0.652 33	0.727 21	0.600 45	1.000 1	0.912 17	0.822 49	0.757 34	1.000 1	0.691 35
Zhihao Liang, Zhihao Li, Songcen Xu, Mingkui Tan, Kui Jia: Instance Segmentation in 3D Scenes using Semantic Superpoint Tree Networks. ICCV2021
TD3D	0.875 10	1.000 1	0.976 22	0.877 12	0.783 27	0.970 1	0.889 1	0.828 18	0.945 5	0.803 20	0.713 22	0.720 22	0.709 17	1.000 1	0.936 12	0.934 3	0.873 15	1.000 1	0.791 16
Maksim Kolodiazhnyi, Anna Vorontsova, Anton Konushin, Danila Rukhovich: Top-Down Beats Bottom-Up in 3D Instance Segmentation. WACV 2024
IPCA-Inst	0.851 18	1.000 1	0.968 26	0.884 11	0.842 11	0.862 37	0.693 22	0.812 23	0.888 19	0.677 34	0.783 5	0.698 23	0.807 7	1.000 1	0.911 24	0.865 33	0.865 17	1.000 1	0.757 24

TopoSeg	0.832 23	1.000 1	0.981 17	0.933 2	0.819 16	0.826 46	0.524 44	0.841 16	0.811 32	0.681 33	0.759 11	0.687 24	0.727 16	0.981 42	0.911 24	0.883 17	0.853 21	1.000 1	0.756 25

PBNet	0.825 25	1.000 1	0.963 28	0.837 26	0.843 10	0.865 32	0.822 4	0.647 48	0.878 21	0.733 24	0.639 37	0.683 25	0.650 28	1.000 1	0.853 38	0.870 29	0.820 26	1.000 1	0.744 26
W.Zhao, Y.Yan, C.Yang, J.Ye,X.Yang,K.Huang: Divide and Conquer: 3D Instance Segmentation With Point-Wise Binarization. ICCV 2023
RPGN	0.806 28	1.000 1	0.992 10	0.789 39	0.723 46	0.891 26	0.650 26	0.810 24	0.832 29	0.665 37	0.699 26	0.658 26	0.700 19	1.000 1	0.881 31	0.832 46	0.774 28	0.997 37	0.613 47
Shichao Dong, Guosheng Lin, Tzu-Yi Hung: Learning Regional Purity for Instance Segmentation on 3D Point Clouds. ECCV 2022
PanopticFusion-inst	0.693 53	1.000 1	0.852 59	0.655 60	0.616 58	0.788 52	0.334 56	0.763 31	0.771 40	0.457 66	0.555 52	0.652 27	0.518 56	0.857 51	0.765 53	0.732 66	0.631 56	0.944 55	0.577 52
Gaku Narita, Takashi Seno, Tomoya Ishikawa, Yohsuke Kaji: PanopticFusion: Online Volumetric Semantic Mapping at the Level of Stuff and Things. IROS 2019 (to appear)
Mask-Group	0.792 31	1.000 1	0.968 27	0.812 31	0.766 32	0.864 33	0.460 47	0.815 22	0.888 18	0.598 47	0.651 34	0.639 28	0.600 45	0.918 48	0.941 7	0.896 12	0.721 42	1.000 1	0.723 29
Min Zhong, Xinghao Chen, Xiaokang Chen, Gang Zeng, Yunhe Wang: MaskGroup: Hierarchical Point Grouping and Masking for 3D Instance Segmentation. ICME 2022
HAIS	0.803 29	1.000 1	0.994 8	0.820 30	0.759 34	0.855 38	0.554 39	0.882 7	0.827 31	0.615 43	0.676 29	0.638 29	0.646 37	1.000 1	0.912 17	0.797 60	0.767 29	0.994 46	0.726 28
Shaoyu Chen, Jiemin Fang, Qian Zhang, Wenyu Liu, Xinggang Wang: Hierarchical Aggregation for 3D Instance Segmentation. ICCV 2021
SSEC	0.820 26	1.000 1	0.983 16	0.924 4	0.826 14	0.817 49	0.415 53	0.899 4	0.793 36	0.673 35	0.731 19	0.636 30	0.653 26	1.000 1	0.939 9	0.804 57	0.878 10	1.000 1	0.780 17

One_Thing_One_Click	0.675 59	1.000 1	0.823 61	0.782 43	0.621 57	0.766 54	0.211 65	0.736 38	0.560 63	0.586 50	0.522 56	0.636 31	0.453 63	0.641 63	0.853 38	0.850 41	0.694 49	0.997 37	0.411 64
Zhengzhe Liu, Xiaojuan Qi, Chi-Wing Fu: One Thing One Click: A Self-Training Approach for Weakly Supervised 3D Semantic Segmentation. CVPR 2021
Mask3D_evaluation	0.843 20	1.000 1	0.955 31	0.847 21	0.795 23	0.932 17	0.750 16	0.780 30	0.891 16	0.818 17	0.737 17	0.633 32	0.703 18	1.000 1	0.902 29	0.870 28	0.820 25	0.941 59	0.805 13

DualGroup	0.782 37	1.000 1	0.927 40	0.811 32	0.772 30	0.853 39	0.631 29	0.805 25	0.773 38	0.613 44	0.611 42	0.610 33	0.650 28	0.835 59	0.881 31	0.879 20	0.750 37	1.000 1	0.675 36

PE	0.776 39	1.000 1	0.900 51	0.860 19	0.728 45	0.869 30	0.400 54	0.857 11	0.774 37	0.568 55	0.701 25	0.602 34	0.646 37	0.933 47	0.843 41	0.890 14	0.691 51	0.997 37	0.709 31
Biao Zhang, Peter Wonka: Point Cloud Instance Segmentation using Probabilistic Embeddings. CVPR 2021
DANCENET	0.786 35	1.000 1	0.936 35	0.783 42	0.737 43	0.852 40	0.742 17	0.647 48	0.765 43	0.811 18	0.624 40	0.579 35	0.632 42	1.000 1	0.909 28	0.898 11	0.696 47	0.944 55	0.601 50

PointGroup	0.778 38	1.000 1	0.900 50	0.798 38	0.715 48	0.863 34	0.493 45	0.706 41	0.895 15	0.569 54	0.701 24	0.576 36	0.639 40	1.000 1	0.880 33	0.851 39	0.719 43	0.997 37	0.709 32
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]
3D-MPA	0.737 46	1.000 1	0.933 37	0.785 40	0.794 25	0.831 44	0.279 61	0.588 54	0.695 51	0.616 42	0.559 51	0.556 37	0.650 28	1.000 1	0.809 47	0.875 24	0.696 48	1.000 1	0.608 49
Francis Engelmann, Martin Bokeloh, Alireza Fathi, Bastian Leibe, Matthias Nießner: 3D-MPA: Multi Proposal Aggregation for 3D Semantic Instance Segmentation. CVPR 2020
GICN	0.788 34	1.000 1	0.978 19	0.867 17	0.781 28	0.833 43	0.527 43	0.824 19	0.806 33	0.549 56	0.596 44	0.551 38	0.700 19	1.000 1	0.853 38	0.935 2	0.733 39	1.000 1	0.651 38

RWSeg	0.739 45	1.000 1	0.899 52	0.759 49	0.753 38	0.823 47	0.282 59	0.691 44	0.658 53	0.582 52	0.594 45	0.547 39	0.628 43	1.000 1	0.795 49	0.868 31	0.728 41	1.000 1	0.692 34

Dyco3D	0.761 43	1.000 1	0.935 36	0.893 8	0.752 39	0.863 35	0.600 30	0.588 54	0.742 45	0.641 39	0.633 38	0.546 40	0.550 52	0.857 51	0.789 52	0.853 38	0.762 32	0.987 49	0.699 33
Tong He; Chunhua Shen; Anton van den Hengel: DyCo3D: Robust Instance Segmentation of 3D Point Clouds through Dynamic Convolution. CVPR2021
AOIA	0.767 40	1.000 1	0.937 34	0.810 33	0.740 42	0.906 21	0.550 40	0.800 27	0.706 48	0.577 53	0.624 39	0.544 41	0.596 50	0.857 51	0.879 35	0.880 19	0.750 36	0.992 47	0.658 37

CSC-Pretrained	0.791 32	1.000 1	0.996 4	0.829 29	0.767 31	0.889 28	0.600 30	0.819 21	0.770 41	0.594 48	0.620 41	0.541 42	0.700 19	1.000 1	0.941 7	0.889 15	0.763 31	1.000 1	0.526 57

DENet	0.786 35	1.000 1	0.929 39	0.736 52	0.750 40	0.720 62	0.755 15	0.934 1	0.794 35	0.590 49	0.561 50	0.537 43	0.650 28	1.000 1	0.882 30	0.804 58	0.789 27	1.000 1	0.719 30

INS-Conv-instance	0.762 42	1.000 1	0.923 43	0.765 47	0.785 26	0.905 22	0.600 30	0.655 47	0.646 55	0.683 32	0.647 35	0.530 44	0.650 28	1.000 1	0.824 43	0.830 47	0.693 50	0.944 55	0.644 40

SphereSeg	0.835 21	1.000 1	0.963 29	0.891 9	0.794 24	0.954 5	0.822 5	0.710 40	0.961 2	0.721 26	0.693 28	0.530 45	0.653 27	1.000 1	0.867 36	0.857 37	0.859 19	0.991 48	0.771 19

MTML	0.731 47	1.000 1	0.992 10	0.779 46	0.609 59	0.746 57	0.308 58	0.867 8	0.601 59	0.607 45	0.539 54	0.519 46	0.550 52	1.000 1	0.824 43	0.869 30	0.729 40	1.000 1	0.616 45
Jean Lahoud, Bernard Ghanem, Marc Pollefeys, Martin R. Oswald: 3D Instance Segmentation via Multi-task Metric Learning. ICCV 2019 [oral]
PCJC	0.684 57	1.000 1	0.895 54	0.757 50	0.659 55	0.862 36	0.189 68	0.739 37	0.606 58	0.712 29	0.581 47	0.515 47	0.650 28	0.857 51	0.357 69	0.785 61	0.631 57	0.889 65	0.635 42

OccuSeg+instance	0.742 44	1.000 1	0.923 43	0.785 40	0.745 41	0.867 31	0.557 36	0.578 57	0.729 46	0.670 36	0.644 36	0.488 48	0.577 51	1.000 1	0.794 50	0.830 47	0.620 60	1.000 1	0.550 53
Lei Han, Tian Zheng, Lan Xu, Lu Fang: OccuSeg: Occupancy-aware 3D Instance Segmentation. CVPR2020
MaskRCNN 2d->3d Proj	0.261 73	0.903 66	0.081 73	0.008 73	0.233 72	0.175 73	0.280 60	0.106 73	0.150 73	0.203 73	0.175 72	0.480 49	0.218 73	0.143 72	0.542 65	0.404 73	0.153 73	0.393 73	0.049 73

Box2Mask	0.803 29	1.000 1	0.962 30	0.874 13	0.707 50	0.887 29	0.686 25	0.598 53	0.961 1	0.715 28	0.694 27	0.469 50	0.700 19	1.000 1	0.912 17	0.902 8	0.753 35	0.997 37	0.637 41
Julian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll: Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes. ECCV 2022
SSEN	0.724 49	1.000 1	0.926 41	0.781 45	0.661 54	0.845 41	0.596 33	0.529 60	0.764 44	0.653 38	0.489 61	0.461 51	0.500 59	0.859 50	0.765 53	0.872 27	0.761 33	1.000 1	0.577 51
Dongsu Zhang, Junha Chun, Sang Kyun Cha, Young Min Kim: Spatial Semantic Embedding Network: Fast 3D Instance Segmentation with Deep Metric Learning. Arxiv
OSIS	0.725 48	1.000 1	0.885 56	0.653 61	0.657 56	0.801 50	0.576 34	0.695 43	0.828 30	0.698 30	0.534 55	0.457 52	0.500 59	0.857 51	0.831 42	0.841 44	0.627 58	1.000 1	0.619 44

DD-UNet+Group	0.764 41	1.000 1	0.897 53	0.837 25	0.753 37	0.830 45	0.459 49	0.824 19	0.699 50	0.629 41	0.653 32	0.438 53	0.650 28	1.000 1	0.880 33	0.858 36	0.690 52	1.000 1	0.650 39
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
MASC	0.615 61	0.711 68	0.802 63	0.540 67	0.757 35	0.777 53	0.029 71	0.577 58	0.588 62	0.521 62	0.600 43	0.436 54	0.534 54	0.697 61	0.616 62	0.838 45	0.526 62	0.980 52	0.534 56
Chen Liu, Yasutaka Furukawa: MASC: Multi-scale Affinity with Sparse Convolution for 3D Instance Segmentation.
ClickSeg_Instance	0.685 56	1.000 1	0.818 62	0.600 64	0.715 49	0.795 51	0.557 36	0.533 59	0.591 61	0.601 46	0.519 57	0.429 55	0.638 41	0.938 46	0.706 58	0.817 52	0.624 59	0.944 55	0.502 59

SegGroup_ins	0.637 60	1.000 1	0.923 45	0.593 65	0.561 63	0.746 58	0.143 70	0.504 62	0.766 42	0.485 64	0.442 62	0.372 56	0.530 55	0.714 60	0.815 46	0.775 62	0.673 54	1.000 1	0.431 63
An Tao, Yueqi Duan, Yi Wei, Jiwen Lu, Jie Zhou: SegGroup: Seg-Level Supervision for 3D Instance and Semantic Segmentation. TIP 2022
Sparse R-CNN	0.714 51	1.000 1	0.926 42	0.694 56	0.699 52	0.890 27	0.636 28	0.516 61	0.693 52	0.743 23	0.588 46	0.369 57	0.601 44	0.594 65	0.800 48	0.886 16	0.676 53	0.986 50	0.546 54

SALoss-ResNet	0.695 52	1.000 1	0.855 58	0.579 66	0.589 61	0.735 60	0.484 46	0.588 54	0.856 25	0.634 40	0.571 48	0.298 58	0.500 59	1.000 1	0.824 43	0.818 50	0.702 46	0.935 62	0.545 55
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)
UNet-backbone	0.605 62	1.000 1	0.909 48	0.764 48	0.603 60	0.704 63	0.415 52	0.301 67	0.548 64	0.461 65	0.394 63	0.267 59	0.386 65	0.857 51	0.649 61	0.817 51	0.504 64	0.959 53	0.356 67

Occipital-SCS	0.688 54	1.000 1	0.913 47	0.730 54	0.737 44	0.743 59	0.442 50	0.855 13	0.655 54	0.546 57	0.546 53	0.263 60	0.508 58	0.889 49	0.568 64	0.771 63	0.705 45	0.889 65	0.625 43

3D-BoNet	0.687 55	1.000 1	0.887 55	0.836 27	0.587 62	0.643 69	0.550 40	0.620 50	0.724 47	0.522 61	0.501 59	0.243 61	0.512 57	1.000 1	0.751 55	0.807 56	0.661 55	0.909 64	0.612 48
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
Sem_Recon_ins	0.484 67	0.764 67	0.608 72	0.470 69	0.521 65	0.637 70	0.311 57	0.218 69	0.348 71	0.365 70	0.223 69	0.222 62	0.258 70	0.629 64	0.734 56	0.596 72	0.509 63	0.858 68	0.444 62

R-PointNet	0.544 64	0.500 71	0.655 70	0.661 59	0.663 53	0.765 55	0.432 51	0.214 70	0.612 57	0.584 51	0.499 60	0.204 63	0.286 69	0.429 68	0.655 60	0.650 71	0.539 61	0.950 54	0.499 60

SPG_WSIS	0.678 58	1.000 1	0.880 57	0.836 27	0.701 51	0.727 61	0.273 63	0.607 52	0.706 49	0.541 59	0.515 58	0.174 64	0.600 45	0.857 51	0.716 57	0.846 43	0.711 44	1.000 1	0.506 58

Sgpn_scannet	0.390 72	0.556 70	0.636 71	0.493 68	0.353 71	0.539 72	0.271 64	0.160 72	0.450 69	0.359 71	0.178 71	0.146 65	0.250 71	0.143 72	0.347 70	0.698 69	0.436 71	0.667 71	0.331 69

tmp	0.474 68	1.000 1	0.727 65	0.433 71	0.481 68	0.673 67	0.022 73	0.380 65	0.517 66	0.436 68	0.338 66	0.128 66	0.343 67	0.429 68	0.291 71	0.728 67	0.473 65	0.833 69	0.300 70

SemRegionNet-20cls	0.470 69	1.000 1	0.727 65	0.447 70	0.481 67	0.678 66	0.024 72	0.380 65	0.518 65	0.440 67	0.339 65	0.128 66	0.350 66	0.429 68	0.212 72	0.711 68	0.465 67	0.833 69	0.290 71

3D-SIS	0.558 63	1.000 1	0.773 64	0.614 63	0.503 66	0.691 65	0.200 66	0.412 63	0.498 67	0.546 58	0.311 68	0.103 68	0.600 45	0.857 51	0.382 66	0.799 59	0.445 70	0.938 61	0.371 65
Ji Hou, Angela Dai, Matthias Niessner: 3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans. CVPR 2019
3D-BEVIS	0.401 71	0.667 69	0.687 69	0.419 72	0.137 73	0.587 71	0.188 69	0.235 68	0.359 70	0.211 72	0.093 73	0.080 69	0.311 68	0.571 66	0.382 66	0.754 65	0.300 72	0.874 67	0.357 66
Cathrin Elich, Francis Engelmann, Jonas Schult, Theodora Kontogianni, Bastian Leibe: 3D-BEVIS: Birds-Eye-View Instance Segmentation.
ASIS	0.422 70	0.333 72	0.707 68	0.676 58	0.401 70	0.650 68	0.350 55	0.177 71	0.594 60	0.376 69	0.202 70	0.077 70	0.404 64	0.571 66	0.197 73	0.674 70	0.447 69	0.500 72	0.260 72

Hier3D	0.540 65	1.000 1	0.727 65	0.626 62	0.467 69	0.693 64	0.200 66	0.412 63	0.480 68	0.528 60	0.318 67	0.077 71	0.600 45	0.688 62	0.382 66	0.768 64	0.472 66	0.941 59	0.350 68
Tan: HCFS3D: Hierarchical Coupled Feature Selection Network for 3D Semantic and Instance Segmentation.
NeuralBF	0.718 50	1.000 1	0.945 33	0.901 7	0.754 36	0.817 48	0.460 47	0.700 42	0.772 39	0.688 31	0.568 49	0.000 72	0.500 59	0.981 42	0.606 63	0.872 26	0.740 38	1.000 1	0.614 46
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
Region-18class	0.497 66	0.250 73	0.902 49	0.689 57	0.540 64	0.747 56	0.276 62	0.610 51	0.268 72	0.489 63	0.348 64	0.000 72	0.243 72	0.220 71	0.663 59	0.814 53	0.459 68	0.928 63	0.496 61

3D Semantic Instance Benchmark