| CutMix (DeepLab v2 ImageNet pre-trained) | 66.48% | Semi-supervised semantic segmentation needs strong, varied perturbations | |
| ReCo (DeepLab v2 with ResNet-101 backbone, ImageNet pretrained) | 68.85% | Bootstrapping Semantic Segmentation with Regional Contrast | |
| Adversarial (DeepLab v2 ImageNet pre-trained) | 59.1% | Adversarial Learning for Semi-Supervised Semantic Segmentation | |
| SemiSegContrast
(DeepLab v2 with ResNet-101 backbone, MSCOCO pretrained) | 70.0% | Semi-Supervised Semantic Segmentation with Pixel-Level Contrastive Learning from a Class-wise Memory Bank | |
| DMT (DeepLab v2 MSCOCO/ImageNet pre-trained) | 69.92% | DMT: Dynamic Mutual Training for Semi-Supervised Learning | |
| s4GAN + MLMT (DeepLab v2 MSCOCO/ImageNet pre-trained) | 67.2% | Semi-Supervised Semantic Segmentation with High- and Low-level Consistency | |
| ClassMix (DeepLab v2 MSCOCO pretrained) | 67.77% | ClassMix: Segmentation-Based Data Augmentation for Semi-Supervised Learning | |
| GIST and RIST (DeepLabv2 with ResNet101, MSCOCO pre-trained) | 69.40% | The GIST and RIST of Iterative Self-Training for Semi-Supervised Segmentation | - |
| Error Localization Network (DeeplabV3 with ResNet-50) | 70.52% | Semi-supervised Semantic Segmentation with Error Localization Network | |
| CutMix (DeepLab v3+ ImageNet pre-trained) | 69.57% | Semi-supervised semantic segmentation needs strong, varied perturbations | |
| ReCo (DeepLab v3+ with ResNet-101 backbone, ImageNet pretrained) | 73.66% | Bootstrapping Semantic Segmentation with Regional Contrast | |
| Error Localization Network (DeeplabV3 with ResNet-101) | 72.52% | Semi-supervised Semantic Segmentation with Error Localization Network | |
| s4GAN+MLMT (DeepLab v3+ ImageNet pre-trained) | 66.6% | Semi-Supervised Semantic Segmentation with High- and Low-level Consistency | |
| s4GAN+MLMT (DeepLab v2 ImageNet pre-trained) | 62.9% | Semi-Supervised Semantic Segmentation with High- and Low-level Consistency | |