From: Ancient mural segmentation based on multiscale feature fusion and dual attention enhancement
Model | Backbone | Animal | Building | Cloud | Disciple | Buddha |
---|---|---|---|---|---|---|
FCN | VGG16 | 0.812 | 0.784 | 0.841 | 0.825 | 0.763 |
SegNet | VGG16 | 0.854 | 0.817 | 0.876 | 0.86 | 0.792 |
DeepLabV3 +  | Xception | 0.873 | 0.834 | 0.915 | 0.871 | 0.817 |
MC-DM | MobileNetV2 | 0.867 | 0.832 | 0.872 | 0.881 | 0.81 |
SETR | ViT-L | 0.88 | 0.853 | 0.876 | 0.864 | 0.825 |
DAFPN | Swin-T | 0.907 | 0.851 | 0.874 | 0.906 | 0.822 |
MFAM (Ours) | CA_MobileViT | 0.935 | 0.898 | 0.875 | 0.915 | 0.852 |