3D Instance Segmentation On Scannetv2

평가 지표

mAP

mAP @ 50

mRec

평가 결과

이 벤치마크에서 각 모델의 성능 결과

모델 이름	mAP	mAP @ 50	mRec	Paper Title	Repository
3D-BoNet	25.3	48.8	47.6	Learning Object Bounding Boxes for 3D Instance Segmentation on Point Clouds
ResNet-Backbone	-	45.9	-	MASC: Multi-scale Affinity with Sparse Convolution for 3D Instance Segmentation
MSTA3D	56.9	79.5	74.1	MSTA3D: Multi-scale Twin-attention for 3D Instance Segmentation	-
RPGN	42.8	64.2	-	Learning Regional Purity for Instance Segmentation on 3D Point Clouds
UNet-Backbone	-	31.9	-	3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation
Searilized Point Mamba	40.0	-	-	Serialized Point Mamba: A Serialized Point Cloud Mamba Segmentation Model	-
HIDA	43.6	63.5	-	HIDA: Towards Holistic Indoor Understanding for the Visually Impaired via Semantic Instance Segmentation with a Wearable Solid-State LiDAR Sensor	-
SSTNet	50.6	69.8	-	Instance Segmentation in 3D Scenes using Semantic Superpoint Tree Networks	-
Spherical Mask	61.6	81.2	-	Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation
PointGroup	40.7	63.6	-	PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation
3D-MPA	35.3	59.1	61.1	3D-MPA: Multi Proposal Aggregation for 3D Semantic Instance Segmentation
MAFT	59.6	-	-	Mask-Attention-Free Transformer for 3D Instance Segmentation
HAIS	45.7	69.9	-	Hierarchical Aggregation for 3D Instance Segmentation
ODIN	50.0	71.0	-	ODIN: A Single Model for 2D and 3D Segmentation
OneFromer3D	56.6	80.1	-	OneFormer3D: One Transformer for Unified Point Cloud Segmentation
ISBNet	55.9	76.3	-	ISBNet: a 3D Point Cloud Instance Segmentation Network with Instance-aware Sampling and Box-aware Dynamic Convolution
BFL	60.6	81.0	-	Beyond the Final Layer: Hierarchical Query Fusion Transformer with Agent-Interpolation Initialization for 3D Instance Segmentation	-
DKNet	53.2	71.8	-	3D Instances as 1D Kernels
TD3D	48.9	75.1	-	Top-Down Beats Bottom-Up in 3D Instance Segmentation
GICN	-	63.8	-	Learning Gaussian Instance Segmentation in Point Clouds

0 of 31 row(s) selected.