| ReCo (DeepLab v3+ with ResNet-101 backbone, ImageNet pretrained) | 72.14% | Bootstrapping Semantic Segmentation with Regional Contrast | |
| CutMix (DeepLab v3+ ImageNet pre-trained) | 67.05% | Semi-supervised semantic segmentation needs strong, varied perturbations | |
| DMT (DeepLab v2 MSCOCO pre-trained) | 67.15% | DMT: Dynamic Mutual Training for Semi-Supervised Learning | |
| Adversarial (DeepLab v2 ImageNet pre-trained) | 49.2% | Adversarial Learning for Semi-Supervised Semantic Segmentation | |
| ReCo (DeepLab v2 with ResNet-101 backbone, ImageNet pretrained) | 66.41% | Bootstrapping Semantic Segmentation with Regional Contrast | |
| s4GAN+MLMT (DeepLab v2 ImageNet pre-trained) | 60.4% | Semi-Supervised Semantic Segmentation with High- and Low-level Consistency | |
| ClassMix (DeepLab v2 MSCOCO pretrained) | 66.15% | ClassMix: Segmentation-Based Data Augmentation for Semi-Supervised Learning | |
| s4GAN+MLMT (DeepLab v3+ ImageNet pre-trained) | 62.6% | Semi-Supervised Semantic Segmentation with High- and Low-level Consistency | |
| CutMix (DeepLab v2 ImageNet pre-trained) | 64.81% | Semi-supervised semantic segmentation needs strong, varied perturbations | |
| s4GAN + MLMT (DeepLab v2 MSCOCO/ImageNet pre-trained) | 63.3% | Semi-Supervised Semantic Segmentation with High- and Low-level Consistency | |
| SemiSegContrast
(DeepLab v2 with ResNet-101 backbone, MSCOCO pretrained) | 67.9% | Semi-Supervised Semantic Segmentation with Pixel-Level Contrastive Learning from a Class-wise Memory Bank | |
| GIST and RIST (DeepLabv2 with ResNet101, MSCOCO pre-trained) | 67.21% | The GIST and RIST of Iterative Self-Training for Semi-Supervised Segmentation | - |