3D Semantic Instance Benchmark
The 3D semantic instance prediction task involves detecting and segmenting the object in an 3D scan mesh.
Evaluation and metricsOur 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 | Info | avg ap 50% | bathtub | bed | bookshelf | cabinet | chair | counter | curtain | desk | door | otherfurniture | picture | refrigerator | shower curtain | sink | sofa | table | toilet | window |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
TST3D | 0.795 4 | 1.000 1 | 0.929 7 | 0.918 4 | 0.709 6 | 0.884 11 | 0.596 1 | 0.704 21 | 0.769 5 | 0.734 6 | 0.644 14 | 0.699 10 | 0.751 10 | 1.000 1 | 0.794 6 | 0.876 4 | 0.757 16 | 0.997 34 | 0.550 25 | |
OneFormer3D | 0.801 2 | 1.000 1 | 0.973 2 | 0.909 5 | 0.698 9 | 0.928 2 | 0.582 2 | 0.668 28 | 0.685 12 | 0.780 2 | 0.687 8 | 0.698 11 | 0.702 12 | 1.000 1 | 0.794 7 | 0.900 2 | 0.784 9 | 0.986 45 | 0.635 4 | |
Maxim Kolodiazhnyi, Anna Vorontsova, Anton Konushin, Danila Rukhovich: OneFormer3D: One Transformer for Unified Point Cloud Segmentation. | ||||||||||||||||||||
Spherical Mask(CtoF) | 0.812 1 | 1.000 1 | 0.973 3 | 0.852 11 | 0.718 3 | 0.917 3 | 0.574 3 | 0.677 25 | 0.748 7 | 0.729 7 | 0.715 4 | 0.795 1 | 0.809 1 | 1.000 1 | 0.831 3 | 0.854 7 | 0.787 7 | 1.000 1 | 0.638 3 | |
SPFormer | 0.770 9 | 0.903 49 | 0.903 9 | 0.806 18 | 0.609 24 | 0.886 8 | 0.568 4 | 0.815 6 | 0.705 11 | 0.711 8 | 0.655 11 | 0.652 17 | 0.685 17 | 1.000 1 | 0.789 9 | 0.809 22 | 0.776 12 | 1.000 1 | 0.583 18 | |
Sun Jiahao, Qing Chunmei, Tan Junpeng, Xu Xiangmin: Superpoint Transformer for 3D Scene Instance Segmentation. AAAI 2023 [Oral] | ||||||||||||||||||||
Queryformer | 0.787 6 | 1.000 1 | 0.933 5 | 0.601 42 | 0.754 1 | 0.886 9 | 0.558 5 | 0.661 30 | 0.767 6 | 0.665 12 | 0.716 3 | 0.639 18 | 0.808 3 | 1.000 1 | 0.844 1 | 0.897 3 | 0.804 4 | 1.000 1 | 0.624 6 | |
SIM3D | 0.766 11 | 1.000 1 | 0.948 4 | 0.582 48 | 0.599 26 | 0.882 12 | 0.510 6 | 0.701 22 | 0.632 19 | 0.772 3 | 0.685 9 | 0.687 13 | 0.782 7 | 1.000 1 | 0.833 2 | 0.756 32 | 0.798 5 | 1.000 1 | 0.622 7 | |
Mask3D | 0.780 8 | 1.000 1 | 0.786 36 | 0.716 32 | 0.696 10 | 0.885 10 | 0.500 7 | 0.714 19 | 0.810 3 | 0.672 11 | 0.715 4 | 0.679 14 | 0.809 1 | 1.000 1 | 0.831 3 | 0.833 14 | 0.787 7 | 1.000 1 | 0.602 13 | |
Jonas Schult, Francis Engelmann, Alexander Hermans, Or Litany, Siyu Tang, Bastian Leibe: Mask3D for 3D Semantic Instance Segmentation. ICRA 2023 | ||||||||||||||||||||
MAFT | 0.786 7 | 1.000 1 | 0.894 12 | 0.807 17 | 0.694 11 | 0.893 7 | 0.486 8 | 0.674 26 | 0.740 8 | 0.786 1 | 0.704 7 | 0.727 4 | 0.739 11 | 1.000 1 | 0.707 17 | 0.849 10 | 0.756 17 | 1.000 1 | 0.685 1 | |
OccuSeg+instance | 0.672 27 | 1.000 1 | 0.758 45 | 0.682 35 | 0.576 30 | 0.842 20 | 0.477 9 | 0.504 51 | 0.524 27 | 0.567 19 | 0.585 27 | 0.451 32 | 0.557 42 | 1.000 1 | 0.751 13 | 0.797 23 | 0.563 43 | 1.000 1 | 0.467 39 | |
Lei Han, Tian Zheng, Lan Xu, Lu Fang: OccuSeg: Occupancy-aware 3D Instance Segmentation. CVPR2020 | ||||||||||||||||||||
ISBNet | 0.757 13 | 1.000 1 | 0.904 8 | 0.731 30 | 0.678 13 | 0.895 5 | 0.458 10 | 0.644 35 | 0.670 14 | 0.710 9 | 0.620 19 | 0.732 3 | 0.650 19 | 1.000 1 | 0.756 12 | 0.778 25 | 0.779 10 | 1.000 1 | 0.614 10 | |
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 | ||||||||||||||||||||
PBNet | 0.747 15 | 1.000 1 | 0.818 21 | 0.837 14 | 0.713 5 | 0.844 19 | 0.457 11 | 0.647 33 | 0.711 10 | 0.614 14 | 0.617 21 | 0.657 16 | 0.650 19 | 1.000 1 | 0.692 18 | 0.822 17 | 0.765 15 | 1.000 1 | 0.595 15 | |
W.Zhao, Y.Yan, C.Yang, J.Ye,X.Yang,K.Huang: Divide and Conquer: 3D Instance Segmentation With Point-Wise Binarization. ICCV 2023 | ||||||||||||||||||||
UniPerception | 0.800 3 | 1.000 1 | 0.930 6 | 0.872 9 | 0.727 2 | 0.862 17 | 0.454 12 | 0.764 13 | 0.820 1 | 0.746 5 | 0.706 6 | 0.750 2 | 0.772 8 | 0.926 38 | 0.764 11 | 0.818 21 | 0.826 1 | 0.997 34 | 0.660 2 | |
ExtMask3D | 0.789 5 | 1.000 1 | 0.988 1 | 0.756 27 | 0.706 7 | 0.912 4 | 0.429 13 | 0.647 33 | 0.806 4 | 0.755 4 | 0.673 10 | 0.689 12 | 0.772 9 | 1.000 1 | 0.789 8 | 0.852 8 | 0.811 3 | 1.000 1 | 0.617 9 | |
TD3D | 0.751 14 | 1.000 1 | 0.774 37 | 0.867 10 | 0.621 20 | 0.934 1 | 0.404 14 | 0.706 20 | 0.812 2 | 0.605 16 | 0.633 17 | 0.626 19 | 0.690 16 | 1.000 1 | 0.640 26 | 0.820 18 | 0.777 11 | 1.000 1 | 0.612 11 | |
Maksim Kolodiazhnyi, Anna Vorontsova, Anton Konushin, Danila Rukhovich: Top-Down Beats Bottom-Up in 3D Instance Segmentation. WACV 2024 | ||||||||||||||||||||
SSTNet | 0.698 22 | 1.000 1 | 0.697 53 | 0.888 7 | 0.556 32 | 0.803 30 | 0.387 15 | 0.626 37 | 0.417 36 | 0.556 21 | 0.585 26 | 0.702 7 | 0.600 34 | 1.000 1 | 0.824 5 | 0.720 41 | 0.692 22 | 1.000 1 | 0.509 30 | |
Zhihao Liang, Zhihao Li, Songcen Xu, Mingkui Tan, Kui Jia: Instance Segmentation in 3D Scenes using Semantic Superpoint Tree Networks. ICCV2021 | ||||||||||||||||||||
DualGroup | 0.694 23 | 1.000 1 | 0.799 31 | 0.811 16 | 0.622 19 | 0.817 24 | 0.376 16 | 0.805 9 | 0.590 23 | 0.487 30 | 0.568 29 | 0.525 27 | 0.650 19 | 0.835 48 | 0.600 31 | 0.829 15 | 0.655 27 | 1.000 1 | 0.526 27 | |
DANCENET | 0.680 24 | 1.000 1 | 0.807 26 | 0.733 29 | 0.600 25 | 0.768 38 | 0.375 17 | 0.543 45 | 0.538 26 | 0.610 15 | 0.599 22 | 0.498 28 | 0.632 32 | 0.981 32 | 0.739 14 | 0.856 6 | 0.633 33 | 0.882 56 | 0.454 40 | |
INS-Conv-instance | 0.657 29 | 1.000 1 | 0.760 43 | 0.667 37 | 0.581 28 | 0.863 15 | 0.323 18 | 0.655 31 | 0.477 29 | 0.473 32 | 0.549 32 | 0.432 36 | 0.650 19 | 1.000 1 | 0.655 22 | 0.738 37 | 0.585 39 | 0.944 48 | 0.472 38 | |
OSIS | 0.605 40 | 1.000 1 | 0.801 30 | 0.599 43 | 0.535 35 | 0.728 44 | 0.286 19 | 0.436 55 | 0.679 13 | 0.491 28 | 0.433 43 | 0.256 48 | 0.404 55 | 0.857 40 | 0.620 29 | 0.724 39 | 0.510 53 | 1.000 1 | 0.539 26 | |
HAIS | 0.699 21 | 1.000 1 | 0.849 15 | 0.820 15 | 0.675 14 | 0.808 29 | 0.279 20 | 0.757 15 | 0.465 31 | 0.517 25 | 0.596 23 | 0.559 22 | 0.600 34 | 1.000 1 | 0.654 23 | 0.767 28 | 0.676 24 | 0.994 41 | 0.560 23 | |
Shaoyu Chen, Jiemin Fang, Qian Zhang, Wenyu Liu, Xinggang Wang: Hierarchical Aggregation for 3D Instance Segmentation. ICCV 2021 | ||||||||||||||||||||
Box2Mask | 0.677 26 | 1.000 1 | 0.847 16 | 0.771 23 | 0.509 41 | 0.816 25 | 0.277 21 | 0.558 44 | 0.482 28 | 0.562 20 | 0.640 15 | 0.448 33 | 0.700 13 | 1.000 1 | 0.666 19 | 0.852 9 | 0.578 40 | 0.997 34 | 0.488 35 | |
Julian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll: Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes. ECCV 2022 | ||||||||||||||||||||
TopoSeg | 0.725 18 | 1.000 1 | 0.806 28 | 0.933 2 | 0.668 15 | 0.758 39 | 0.272 22 | 0.734 18 | 0.630 20 | 0.549 22 | 0.654 12 | 0.606 20 | 0.697 15 | 0.966 35 | 0.612 30 | 0.839 12 | 0.754 18 | 1.000 1 | 0.573 19 | |
IPCA-Inst | 0.731 17 | 1.000 1 | 0.788 35 | 0.884 8 | 0.698 8 | 0.788 35 | 0.252 23 | 0.760 14 | 0.646 18 | 0.511 26 | 0.637 16 | 0.665 15 | 0.804 5 | 1.000 1 | 0.644 25 | 0.778 26 | 0.747 19 | 1.000 1 | 0.561 22 | |
GraphCut | 0.732 16 | 1.000 1 | 0.788 34 | 0.724 31 | 0.642 18 | 0.859 18 | 0.248 24 | 0.787 11 | 0.618 21 | 0.596 17 | 0.653 13 | 0.722 6 | 0.583 40 | 1.000 1 | 0.766 10 | 0.861 5 | 0.825 2 | 1.000 1 | 0.504 31 | |
SoftGroup | 0.761 12 | 1.000 1 | 0.808 25 | 0.845 12 | 0.716 4 | 0.862 16 | 0.243 25 | 0.824 4 | 0.655 17 | 0.620 13 | 0.734 2 | 0.699 9 | 0.791 6 | 0.981 32 | 0.716 15 | 0.844 11 | 0.769 13 | 1.000 1 | 0.594 16 | |
Thang Vu, Kookhoi Kim, Tung M. Luu, Xuan Thanh Nguyen, Chang D. Yoo: SoftGroup for 3D Instance Segmentaiton on Point Clouds. CVPR 2022 [Oral] | ||||||||||||||||||||
DKNet | 0.718 19 | 1.000 1 | 0.814 22 | 0.782 21 | 0.619 21 | 0.872 13 | 0.224 26 | 0.751 16 | 0.569 25 | 0.677 10 | 0.585 25 | 0.724 5 | 0.633 30 | 0.981 32 | 0.515 40 | 0.819 19 | 0.736 20 | 1.000 1 | 0.617 8 | |
Yizheng Wu, Min Shi, Shuaiyuan Du, Hao Lu, Zhiguo Cao, Weicai Zhong: 3D Instances as 1D Kernels. ECCV 2022 | ||||||||||||||||||||
DENet | 0.629 38 | 1.000 1 | 0.797 32 | 0.608 40 | 0.589 27 | 0.627 51 | 0.219 27 | 0.882 1 | 0.310 45 | 0.402 48 | 0.383 51 | 0.396 41 | 0.650 19 | 1.000 1 | 0.663 21 | 0.543 59 | 0.691 23 | 1.000 1 | 0.568 21 | |
SoftGroup++ | 0.769 10 | 1.000 1 | 0.803 29 | 0.937 1 | 0.684 12 | 0.865 14 | 0.213 28 | 0.870 2 | 0.664 15 | 0.571 18 | 0.758 1 | 0.702 8 | 0.807 4 | 1.000 1 | 0.653 24 | 0.902 1 | 0.792 6 | 1.000 1 | 0.626 5 | |
Sparse R-CNN | 0.515 49 | 1.000 1 | 0.538 61 | 0.282 55 | 0.468 47 | 0.790 34 | 0.173 29 | 0.345 57 | 0.429 34 | 0.413 46 | 0.484 38 | 0.176 52 | 0.595 39 | 0.591 56 | 0.522 39 | 0.668 50 | 0.476 54 | 0.986 46 | 0.327 51 | |
SSEC | 0.707 20 | 1.000 1 | 0.850 14 | 0.924 3 | 0.648 16 | 0.747 42 | 0.162 30 | 0.862 3 | 0.572 24 | 0.520 24 | 0.624 18 | 0.549 23 | 0.649 28 | 1.000 1 | 0.560 35 | 0.706 42 | 0.768 14 | 1.000 1 | 0.591 17 | |
SphereSeg | 0.680 24 | 1.000 1 | 0.856 13 | 0.744 28 | 0.618 22 | 0.893 6 | 0.151 31 | 0.651 32 | 0.713 9 | 0.537 23 | 0.579 28 | 0.430 37 | 0.651 18 | 1.000 1 | 0.389 51 | 0.744 36 | 0.697 21 | 0.991 43 | 0.601 14 | |
DD-UNet+Group | 0.635 36 | 0.667 51 | 0.797 33 | 0.714 33 | 0.562 31 | 0.774 37 | 0.146 32 | 0.810 8 | 0.429 35 | 0.476 31 | 0.546 34 | 0.399 40 | 0.633 30 | 1.000 1 | 0.632 27 | 0.722 40 | 0.609 34 | 1.000 1 | 0.514 28 | |
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 | ||||||||||||||||||||
GICN | 0.638 34 | 1.000 1 | 0.895 11 | 0.800 19 | 0.480 45 | 0.676 47 | 0.144 33 | 0.737 17 | 0.354 42 | 0.447 35 | 0.400 49 | 0.365 44 | 0.700 13 | 1.000 1 | 0.569 33 | 0.836 13 | 0.599 35 | 1.000 1 | 0.473 37 | |
CSC-Pretrained | 0.648 30 | 1.000 1 | 0.810 23 | 0.768 24 | 0.523 39 | 0.813 27 | 0.143 34 | 0.819 5 | 0.389 39 | 0.422 43 | 0.511 36 | 0.443 34 | 0.650 19 | 1.000 1 | 0.624 28 | 0.732 38 | 0.634 32 | 1.000 1 | 0.375 47 | |
Mask-Group | 0.664 28 | 1.000 1 | 0.822 20 | 0.764 26 | 0.616 23 | 0.815 26 | 0.139 35 | 0.694 24 | 0.597 22 | 0.459 34 | 0.566 30 | 0.599 21 | 0.600 34 | 0.516 58 | 0.715 16 | 0.819 20 | 0.635 31 | 1.000 1 | 0.603 12 | |
Min Zhong, Xinghao Chen, Xiaokang Chen, Gang Zeng, Yunhe Wang: MaskGroup: Hierarchical Point Grouping and Masking for 3D Instance Segmentation. ICME 2022 | ||||||||||||||||||||
SALoss-ResNet | 0.459 54 | 1.000 1 | 0.737 47 | 0.159 65 | 0.259 57 | 0.587 54 | 0.138 36 | 0.475 53 | 0.217 51 | 0.416 45 | 0.408 47 | 0.128 54 | 0.315 58 | 0.714 51 | 0.411 48 | 0.536 60 | 0.590 38 | 0.873 57 | 0.304 52 | |
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) | ||||||||||||||||||||
3D-MPA | 0.611 39 | 1.000 1 | 0.833 18 | 0.765 25 | 0.526 38 | 0.756 40 | 0.136 37 | 0.588 42 | 0.470 30 | 0.438 39 | 0.432 45 | 0.358 46 | 0.650 19 | 0.857 40 | 0.429 47 | 0.765 29 | 0.557 46 | 1.000 1 | 0.430 42 | |
Francis Engelmann, Martin Bokeloh, Alireza Fathi, Bastian Leibe, Matthias Nießner: 3D-MPA: Multi Proposal Aggregation for 3D Semantic Instance Segmentation. CVPR 2020 | ||||||||||||||||||||
ClickSeg_Instance | 0.539 47 | 1.000 1 | 0.621 56 | 0.300 54 | 0.530 37 | 0.698 45 | 0.127 38 | 0.533 46 | 0.222 50 | 0.430 42 | 0.400 48 | 0.365 44 | 0.574 41 | 0.938 36 | 0.472 43 | 0.659 51 | 0.543 49 | 0.944 48 | 0.347 50 | |
PointGroup | 0.636 35 | 1.000 1 | 0.765 40 | 0.624 39 | 0.505 43 | 0.797 32 | 0.116 39 | 0.696 23 | 0.384 40 | 0.441 36 | 0.559 31 | 0.476 30 | 0.596 37 | 1.000 1 | 0.666 19 | 0.756 33 | 0.556 47 | 0.997 34 | 0.513 29 | |
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] | ||||||||||||||||||||
Dyco3D | 0.641 33 | 1.000 1 | 0.841 17 | 0.893 6 | 0.531 36 | 0.802 31 | 0.115 40 | 0.588 42 | 0.448 33 | 0.438 38 | 0.537 35 | 0.430 38 | 0.550 43 | 0.857 40 | 0.534 38 | 0.764 30 | 0.657 26 | 0.987 44 | 0.568 20 | |
Tong He; Chunhua Shen; Anton van den Hengel: DyCo3D: Robust Instance Segmentation of 3D Point Clouds through Dynamic Convolution. CVPR2021 | ||||||||||||||||||||
3D-BoNet | 0.488 51 | 1.000 1 | 0.672 55 | 0.590 45 | 0.301 55 | 0.484 62 | 0.098 41 | 0.620 38 | 0.306 46 | 0.341 54 | 0.259 57 | 0.125 55 | 0.434 52 | 0.796 50 | 0.402 49 | 0.499 61 | 0.513 52 | 0.909 52 | 0.439 41 | |
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 | ||||||||||||||||||||
Occipital-SCS | 0.512 50 | 1.000 1 | 0.716 49 | 0.509 50 | 0.506 42 | 0.611 52 | 0.092 42 | 0.602 41 | 0.177 53 | 0.346 53 | 0.383 50 | 0.165 53 | 0.442 50 | 0.850 47 | 0.386 52 | 0.618 55 | 0.543 50 | 0.889 53 | 0.389 46 | |
PE | 0.645 31 | 1.000 1 | 0.773 39 | 0.798 20 | 0.538 34 | 0.786 36 | 0.088 43 | 0.799 10 | 0.350 43 | 0.435 41 | 0.547 33 | 0.545 24 | 0.646 29 | 0.933 37 | 0.562 34 | 0.761 31 | 0.556 48 | 0.997 34 | 0.501 33 | |
Biao Zhang, Peter Wonka: Point Cloud Instance Segmentation using Probabilistic Embeddings. CVPR 2021 | ||||||||||||||||||||
Mask3D_evaluation | 0.631 37 | 1.000 1 | 0.829 19 | 0.606 41 | 0.646 17 | 0.836 21 | 0.068 44 | 0.511 49 | 0.462 32 | 0.507 27 | 0.619 20 | 0.389 42 | 0.610 33 | 1.000 1 | 0.432 46 | 0.828 16 | 0.673 25 | 0.788 60 | 0.552 24 | |
SSEN | 0.575 43 | 1.000 1 | 0.761 41 | 0.473 51 | 0.477 46 | 0.795 33 | 0.066 45 | 0.529 47 | 0.658 16 | 0.460 33 | 0.461 41 | 0.380 43 | 0.331 57 | 0.859 39 | 0.401 50 | 0.692 48 | 0.653 28 | 1.000 1 | 0.348 49 | |
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 44 | 0.528 61 | 0.708 52 | 0.626 38 | 0.580 29 | 0.745 43 | 0.063 46 | 0.627 36 | 0.240 49 | 0.400 49 | 0.497 37 | 0.464 31 | 0.515 44 | 1.000 1 | 0.475 42 | 0.745 35 | 0.571 41 | 1.000 1 | 0.429 43 | |
R-PointNet | 0.306 60 | 0.500 63 | 0.405 65 | 0.311 53 | 0.348 52 | 0.589 53 | 0.054 47 | 0.068 65 | 0.126 57 | 0.283 57 | 0.290 55 | 0.028 63 | 0.219 61 | 0.214 62 | 0.331 54 | 0.396 65 | 0.275 61 | 0.821 59 | 0.245 56 | |
RPGN | 0.643 32 | 1.000 1 | 0.758 44 | 0.582 49 | 0.539 33 | 0.826 23 | 0.046 48 | 0.765 12 | 0.372 41 | 0.436 40 | 0.588 24 | 0.539 26 | 0.650 19 | 1.000 1 | 0.577 32 | 0.750 34 | 0.653 29 | 0.997 34 | 0.495 34 | |
Shichao Dong, Guosheng Lin, Tzu-Yi Hung: Learning Regional Purity for Instance Segmentation on 3D Point Clouds. ECCV 2022 | ||||||||||||||||||||
PCJC | 0.578 42 | 1.000 1 | 0.810 24 | 0.583 47 | 0.449 48 | 0.813 28 | 0.042 49 | 0.603 40 | 0.341 44 | 0.490 29 | 0.465 40 | 0.410 39 | 0.650 19 | 0.835 48 | 0.264 57 | 0.694 46 | 0.561 44 | 0.889 53 | 0.504 32 | |
NeuralBF | 0.555 45 | 0.667 51 | 0.896 10 | 0.843 13 | 0.517 40 | 0.751 41 | 0.029 50 | 0.519 48 | 0.414 37 | 0.439 37 | 0.465 39 | 0.000 67 | 0.484 46 | 0.857 40 | 0.287 55 | 0.693 47 | 0.651 30 | 1.000 1 | 0.485 36 | |
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 | ||||||||||||||||||||
Sgpn_scannet | 0.143 67 | 0.208 68 | 0.390 66 | 0.169 63 | 0.065 66 | 0.275 67 | 0.029 51 | 0.069 64 | 0.000 67 | 0.087 67 | 0.043 66 | 0.014 65 | 0.027 68 | 0.000 67 | 0.112 63 | 0.351 67 | 0.168 67 | 0.438 66 | 0.138 65 | |
3D-BEVIS | 0.248 63 | 0.667 51 | 0.566 58 | 0.076 66 | 0.035 68 | 0.394 66 | 0.027 52 | 0.035 67 | 0.098 59 | 0.099 66 | 0.030 67 | 0.025 64 | 0.098 64 | 0.375 61 | 0.126 61 | 0.604 57 | 0.181 66 | 0.854 58 | 0.171 61 | |
Cathrin Elich, Francis Engelmann, Jonas Schult, Theodora Kontogianni, Bastian Leibe: 3D-BEVIS: Birds-Eye-View Instance Segmentation. | ||||||||||||||||||||
Sem_Recon_ins | 0.227 65 | 0.764 50 | 0.486 62 | 0.069 67 | 0.098 65 | 0.426 65 | 0.017 53 | 0.067 66 | 0.015 63 | 0.172 62 | 0.100 62 | 0.096 57 | 0.054 67 | 0.183 63 | 0.135 60 | 0.366 66 | 0.260 65 | 0.614 64 | 0.168 62 | |
3D-SIS | 0.382 57 | 1.000 1 | 0.432 64 | 0.245 57 | 0.190 59 | 0.577 55 | 0.013 54 | 0.263 59 | 0.033 62 | 0.320 55 | 0.240 58 | 0.075 58 | 0.422 53 | 0.857 40 | 0.117 62 | 0.699 44 | 0.271 63 | 0.883 55 | 0.235 57 | |
Ji Hou, Angela Dai, Matthias Niessner: 3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans. CVPR 2019 | ||||||||||||||||||||
UNet-backbone | 0.319 59 | 0.667 51 | 0.715 50 | 0.233 58 | 0.189 60 | 0.479 63 | 0.008 55 | 0.218 60 | 0.067 61 | 0.201 61 | 0.173 60 | 0.107 56 | 0.123 63 | 0.438 59 | 0.150 59 | 0.615 56 | 0.355 56 | 0.916 51 | 0.093 67 | |
AOIA | 0.601 41 | 1.000 1 | 0.761 42 | 0.687 34 | 0.485 44 | 0.828 22 | 0.008 55 | 0.663 29 | 0.405 38 | 0.405 47 | 0.425 46 | 0.490 29 | 0.596 37 | 0.714 51 | 0.553 37 | 0.779 24 | 0.597 36 | 0.992 42 | 0.424 44 | |
MTML | 0.549 46 | 1.000 1 | 0.807 27 | 0.588 46 | 0.327 53 | 0.647 49 | 0.004 57 | 0.815 7 | 0.180 52 | 0.418 44 | 0.364 53 | 0.182 51 | 0.445 49 | 1.000 1 | 0.442 45 | 0.688 49 | 0.571 42 | 1.000 1 | 0.396 45 | |
Jean Lahoud, Bernard Ghanem, Marc Pollefeys, Martin R. Oswald: 3D Instance Segmentation via Multi-task Metric Learning. ICCV 2019 [oral] | ||||||||||||||||||||
MASC | 0.447 55 | 0.528 61 | 0.555 59 | 0.381 52 | 0.382 50 | 0.633 50 | 0.002 58 | 0.509 50 | 0.260 47 | 0.361 52 | 0.432 44 | 0.327 47 | 0.451 48 | 0.571 57 | 0.367 53 | 0.639 53 | 0.386 55 | 0.980 47 | 0.276 54 | |
Chen Liu, Yasutaka Furukawa: MASC: Multi-scale Affinity with Sparse Convolution for 3D Instance Segmentation. | ||||||||||||||||||||
MaskRCNN 2d->3d Proj | 0.058 68 | 0.333 64 | 0.002 68 | 0.000 68 | 0.053 67 | 0.002 68 | 0.002 59 | 0.021 68 | 0.000 67 | 0.045 68 | 0.024 68 | 0.238 49 | 0.065 66 | 0.000 67 | 0.014 67 | 0.107 68 | 0.020 68 | 0.110 68 | 0.006 68 | |
One_Thing_One_Click | 0.529 48 | 0.667 51 | 0.718 48 | 0.777 22 | 0.399 49 | 0.683 46 | 0.000 60 | 0.669 27 | 0.138 55 | 0.391 50 | 0.374 52 | 0.539 25 | 0.360 56 | 0.641 55 | 0.556 36 | 0.774 27 | 0.593 37 | 0.997 34 | 0.251 55 | |
Zhengzhe Liu, Xiaojuan Qi, Chi-Wing Fu: One Thing One Click: A Self-Training Approach for Weakly Supervised 3D Semantic Segmentation. CVPR 2021 | ||||||||||||||||||||
Region-18class | 0.284 61 | 0.250 67 | 0.751 46 | 0.228 60 | 0.270 56 | 0.521 59 | 0.000 60 | 0.468 54 | 0.008 66 | 0.205 60 | 0.127 61 | 0.000 67 | 0.068 65 | 0.070 66 | 0.262 58 | 0.652 52 | 0.323 58 | 0.740 61 | 0.173 60 | |
SegGroup_ins | 0.445 56 | 0.667 51 | 0.773 38 | 0.185 62 | 0.317 54 | 0.656 48 | 0.000 60 | 0.407 56 | 0.134 56 | 0.381 51 | 0.267 56 | 0.217 50 | 0.476 47 | 0.714 51 | 0.452 44 | 0.629 54 | 0.514 51 | 1.000 1 | 0.222 58 | |
An Tao, Yueqi Duan, Yi Wei, Jiwen Lu, Jie Zhou: SegGroup: Seg-Level Supervision for 3D Instance and Semantic Segmentation. TIP 2022 | ||||||||||||||||||||
SPG_WSIS | 0.470 53 | 0.667 51 | 0.685 54 | 0.677 36 | 0.372 51 | 0.562 56 | 0.000 60 | 0.482 52 | 0.244 48 | 0.316 56 | 0.298 54 | 0.052 62 | 0.442 51 | 0.857 40 | 0.267 56 | 0.702 43 | 0.559 45 | 1.000 1 | 0.287 53 | |
SemRegionNet-20cls | 0.250 62 | 0.333 64 | 0.613 57 | 0.229 59 | 0.163 61 | 0.493 60 | 0.000 60 | 0.304 58 | 0.107 58 | 0.147 64 | 0.100 63 | 0.052 61 | 0.231 59 | 0.119 64 | 0.039 64 | 0.445 63 | 0.325 57 | 0.654 62 | 0.141 63 | |
tmp | 0.248 63 | 0.667 51 | 0.437 63 | 0.188 61 | 0.153 63 | 0.491 61 | 0.000 60 | 0.208 61 | 0.094 60 | 0.153 63 | 0.099 64 | 0.057 60 | 0.217 62 | 0.119 64 | 0.039 64 | 0.466 62 | 0.302 59 | 0.640 63 | 0.140 64 | |
ASIS | 0.199 66 | 0.333 64 | 0.253 67 | 0.167 64 | 0.140 64 | 0.438 64 | 0.000 60 | 0.177 63 | 0.008 65 | 0.121 65 | 0.069 65 | 0.004 66 | 0.231 60 | 0.429 60 | 0.036 66 | 0.445 64 | 0.273 62 | 0.333 67 | 0.119 66 | |
Hier3D | 0.323 58 | 0.667 51 | 0.542 60 | 0.264 56 | 0.157 62 | 0.550 57 | 0.000 60 | 0.205 62 | 0.009 64 | 0.270 58 | 0.218 59 | 0.075 58 | 0.500 45 | 0.688 54 | 0.007 68 | 0.698 45 | 0.301 60 | 0.459 65 | 0.200 59 | |
Tan: HCFS3D: Hierarchical Coupled Feature Selection Network for 3D Semantic and Instance Segmentation. | ||||||||||||||||||||
PanopticFusion-inst | 0.478 52 | 0.667 51 | 0.712 51 | 0.595 44 | 0.259 58 | 0.550 58 | 0.000 60 | 0.613 39 | 0.175 54 | 0.250 59 | 0.434 42 | 0.437 35 | 0.411 54 | 0.857 40 | 0.485 41 | 0.591 58 | 0.267 64 | 0.944 48 | 0.359 48 | |
Gaku Narita, Takashi Seno, Tomoya Ishikawa, Yohsuke Kaji: PanopticFusion: Online Volumetric Semantic Mapping at the Level of Stuff and Things. IROS 2019 (to appear) |