全景分割
全景分割是计算机视觉领域的一项任务,旨在结合语义分割和实例分割,以提供场景的全面理解。其目标是将图像分割成具有语义意义的部分或区域,并检测和区分这些区域内的各个对象实例。每个像素都被分配一个语义标签,而属于“事物”类的像素(如可计数的对象实例)则被赋予唯一的实例ID。
COCO test-dev
Mask DINO (single scale)
Cityscapes val
Panoptic FCN* (Swin-L, Cityscapes-fine)
COCO minival
OpenSeeD (SwinL, single-scale)
ADE20K val
DiNAT-L (Mask2Former, 640x640)
Mapillary val
OneFormer (DiNAT-L, single-scale)
Cityscapes test
EfficientPS
LaRS
Mask2Former (Swin-B)
S3DIS Area5
ScanNetV2
OneFormer3D
Indian Driving Dataset
EfficientPS
KITTI Panoptic Segmentation
EfficientPS
PanNuke
LKCell
ScanNet
OneFormer3D
PASTIS
Exchanger+Mask2Former
COCO panoptic
VAN-B6*
MUSES: MUlti-SEnsor Semantic perception dataset
NYU Depth v2
SemanticKITTI
P3Former
ADE20K
MasQCLIP
DALES
SuperCluster
Hypersim
KITTI-360
Panoptic nuScenes val
Panoptic nuScenes test
(AF)2-S3Net + CenterPoint
PASTIS-R
Early Fusion
S3DIS
SUN-RGBD