Semantic Segmentation On Pascal Voc 2012 Val

评估指标

mIoU

评测结果

各个模型在此基准测试上的表现结果

		Paper Title	Repository
EfficientNet-L2+NAS-FPN (single scale test, with self-training)	90.0%	Rethinking Pre-training and Self-training
TADP	87.11%	Text-image Alignment for Diffusion-based Perception
Eff-B7 NAS-FPN (Copy-Paste pre-training, single-scale))	86.6%	Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation
ExFuse (ResNeXt-131)	85.8%	ExFuse: Enhancing Feature Fusion for Semantic Segmentation	-
SpineNet-S143 (single-scale test)	85.64%	Dilated SpineNet for Semantic Segmentation	-
DeepLabv3-JFT	82.7%	Rethinking Atrous Convolution for Semantic Image Segmentation
Auto-DeepLab-L	82.04%	Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Image Segmentation
ResNet-GCN	81.0%	Large Kernel Matters -- Improve Semantic Segmentation by Global Convolutional Network
HyperSeg-L	80.61%	HyperSeg: Patch-wise Hypernetwork for Real-time Semantic Segmentation
DFN (ResNet-101)	80.60%	Learning a Discriminative Feature Network for Semantic Segmentation
WASPnet-CRF (ours)	80.41%	Waterfall Atrous Spatial Pooling Architecture for Efficient Semantic Segmentation
Deeplab v3+ (Res2Net-101)	79.3%	Res2Net: A New Multi-scale Backbone Architecture
FastDenseNas-arch0	78.0%	Fast Neural Architecture Search of Compact Semantic Segmentation Models via Auxiliary Cells
ReLICv2	77.9%	Pushing the limits of self-supervised ResNets: Can we outperform supervised learning without labels on ImageNet?
DeepLab-CRF (ResNet-101)	77.69%	DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs
FastDenseNas-arch2	77.3%	Fast Neural Architecture Search of Compact Semantic Segmentation Models via Auxiliary Cells
DetCon	77.3%	Pushing the limits of self-supervised ResNets: Can we outperform supervised learning without labels on ImageNet?
FastDenseNas-arch1	77.1%	Fast Neural Architecture Search of Compact Semantic Segmentation Models via Auxiliary Cells
DeepLabv3 (ImageNet+300M)	76.5%	Revisiting Unreasonable Effectiveness of Data in Deep Learning Era
BYOL	75.7%	Pushing the limits of self-supervised ResNets: Can we outperform supervised learning without labels on ImageNet?

0 of 29 row(s) selected.