Semantic Segmentation On Cityscapes

评估指标

Mean IoU (class)

评测结果

各个模型在此基准测试上的表现结果

		Paper Title	Repository
VLTSeg	86.4	Strong but simple: A Baseline for Domain Generalized Dense Perception by CLIP-based Transfer Learning
MetaPrompt-SD	86.2	Harnessing Diffusion Models for Visual Perception with Meta Prompts
InternImage-H	86.1%	InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions
HS3-Fuse	85.8%	HS3: Learning with Proper Task Complexity in Hierarchically Supervised Semantic Segmentation	-
InverseForm	85.6%	InverseForm: A Loss Function for Structured Boundary-Aware Segmentation
ViT-Adapter-L (Mask2Former, BEiT pretrain)	85.2%	Vision Transformer Adapter for Dense Predictions
SERNet-Former	84.83	SERNet-Former: Semantic Segmentation by Efficient Residual Network with Attention-Boosting Gates and Attention-Fusion Networks
Depth Anything	84.8%	Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data
HRNetV2 + OCR +	84.5%	Segmentation Transformer: Object-Contextual Representations for Semantic Segmentation
EfficientPS	84.21%	EfficientPS: Efficient Panoptic Segmentation
Panoptic-DeepLab	84.2%	Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation
HRNetV2 + OCR (w/ ASP)	83.7%	Segmentation Transformer: Object-Contextual Representations for Semantic Segmentation
DCNAS(coarse + Mapillary)	83.6%	DCNAS: Densely Connected Neural Architecture Search for Semantic Image Segmentation	-
Euclidean Frank-Wolfe CRFs (backbone: DeepLabv3+)(coarse)	83.6%	Regularized Frank-Wolfe for Dense CRFs: Generalizing Mean Field and Beyond
ResNeSt200 (Mapillary)	83.3%	ResNeSt: Split-Attention Networks
GALDNet(+Mapillary)++	83.3%	Global Aggregation then Local Distribution in Fully Convolutional Networks
HANet (Height-driven Attention Networks by LGE A&B)(coarse)	83.2%	Cars Can't Fly up in the Sky: Improving Urban-Scene Segmentation via Height-driven Attention Networks
kMaX-DeepLab (ConvNeXt-L, fine only)	83.2%	kMaX-DeepLab: k-means Mask Transformer
SegFormer (MiT-B5, Mapillary)	83.1%	SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers
MRFM(coarse)	83.0%	Multi Receptive Field Network for Semantic Segmentation	-

0 of 105 row(s) selected.