Semi-Supervised Semantic Segmentation
Semi-Supervised Semantic Segmentation is a task in the field of computer vision that aims to train models using a small amount of labeled data and a large amount of unlabeled data to achieve the goal of classifying each pixel in an image. This method effectively leverages unlabeled data to improve the model's generalization ability and segmentation accuracy, reducing annotation costs and enhancing the model's application value in real-world scenarios.
Pascal VOC 2012 12.5% labeled
CPS
Cityscapes 12.5% labeled
CPS (DeepLab v3+ with ImageNet-pretrained ResNet-101, single scale inference)
Cityscapes 25% labeled
CPS (DeepLab v3+ with ImageNet-pretrained ResNet-101, single scale inference)
PASCAL VOC 2012 25% labeled
Dual Teacher
Cityscapes 50% labeled
ClassMix (DeepLab v2 MSCOCO pretrained)
Cityscapes 6.25% labeled
Pascal VOC 2012 6.25% labeled
PASCAL VOC 2012 1464 labels
Pascal VOC 2012 5% labeled
ReCo (DeepLab v3+ with ResNet-101 backbone, ImageNet pretrained)
PASCAL VOC 2012 732 labeled
SemiVL (ViT-B/16)
PASCAL VOC 2012 92 labeled
PASCAL VOC 2012 50%
S4MC
Cityscapes 100 samples labeled
ClassMix (DeepLab v2 MSCOCO pretrained)
PASCAL VOC 2012 366 labeled
SemiVL (ViT-B/16)
PASCAL VOC 2012 183 labeled
SemanticKITTI
Pascal VOC 2012 2% labeled
ReCo (DeepLab v3+ with ResNet-101 backbone, ImageNet pretrained)
nuScenes
PLE (Voxel)
COCO 1/64 labeled
COCO 1/256 labeled
ScribbleKITTI
LaserMix (Voxel)
COCO 1/128 labeled
COCO 1/512 labeled
SemiVL
COCO 1/32 labeled
Pascal VOC 2012 1% labeled
ReCo (DeepLab v3+ with ResNet-101 backbone, ImageNet pre-trained)
ADE20K 1/32 labeled
ADE20K 1/16 labeled
UniMatch V2
Cityscapes 93 labeled
AEL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K)
Cityscapes 5% labeled
Cityscapes 2% labeled
WoodScape
FishSegSSL
PASCAL Context 12.5% labeled
GuidedMix-Net(DeepLab v2 with ResNet101, ImageNet pretrained)
Cityscapes with extra (no coarse labels)
Dense FixMatch (DeepLabv3+ ResNet-101, over-sampling, single pass eval)
PASCAL Context 25% labeled
GuidedMix-Net(DeepLab v2 with ResNet101, ImageNet pretrained)
Pascal VOC 2012 50% labeled
AllSpark
Stanford 2D-3D
Cityscapes 10% labeled
IM++ (416x208, 2.7m parameters, no pretraining)
Kvasir-Instrument
MMS(20% labeled)
2017 Robotic Instrument Segmentation Challenge
MMS (20% Labeled)
PASCAL VOC 2012 500 labels
GuidedMix-Net(DeepLab v2 with ResNet50, ImageNet pretrained)
PASCAL VOC 2012 1000 labels
GuidedMix-Net(DeepLab v2 with ResNet50, ImageNet pretrained)
PASCAL VOC 2012 331 labeled
AEL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K)
2D-3D-S
M3L (Linear Fusion B2)
KiTS19
SUIM