Semantic Segmentation on COCO-Stuff (unseen)
61.5mIoUZegFormer
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ZegFormerbackbone=R-50, class embed.=clip text, setting=ZS32021.12 | 61.5 | — | — | |
| ZegFormer-segbackbone=R-50, class embed.=clip text, setting=ZS32021.12 | 48.8 | — | — | |
| SPNet-FPNbackbone=R-50, class embed.=clip text, setting=ZS32021.12 | 41.3 | — | — | |
| Fully Sup.Backbone=DeepLabv2-ResNet1012021.12 | 39.9 | — | — | |
| MaskCLIP+Learning Paradigm=Transductive, Backbone=DeepLabv2-ResNet101, Pseudo Label Backbone=CLIP-ResNet-502021.12 | 39.6 | 5,470 | 45 | |
| SPNetbackbone=R-101, class embed.=ft+w2v, setting=ZS32021.12 | 35.2 | — | — | |
| STRICTLearning Paradigm=Transductive, Backbone=DeepLabv2-ResNet1012021.12 | 34.9 | 3,030 | 32.6 | |
| SPNet+STLearning Paradigm=Transductive, Self-training=true, Backbone=DeepLabv2-ResNet1012021.12 | 34 | 2,690 | 30.3 | |
| ZS3Net+STLearning Paradigm=Transductive, Self-training=true, Backbone=DeepLabv2-ResNet1012021.12 | 33.7 | 1,060 | 16.2 | |
| CaGNet+STLearning Paradigm=Transductive, Self-training=true, Backbone=DeepLabv2-ResNet1012021.12 | 33.7 | 1,340 | 19.5 | |
| CaGNetLearning Paradigm=Inductive, Backbone=DeepLabv2-ResNet1012021.12 | 33.5 | 1,220 | 18.2 | |
| ZS3NetLearning Paradigm=Inductive, Backbone=DeepLabv2-ResNet1012021.12 | 33.3 | 950 | 15 | |
| SPNet-CLearning Paradigm=Inductive, Calibration=true, Backbone=DeepLabv2-ResNet1012021.12 | 32.8 | 870 | 14 | |
| SPNetLearning Paradigm=Inductive, Backbone=DeepLabv2-ResNet1012021.12 | 31.6 | 70 | 1.4 |