From: Ancient mural segmentation based on multiscale feature fusion and dual attention enhancement
Model | Backbone | MIoU/% | MPA/% | Parameters/M |
---|---|---|---|---|
SegNet | VGG16 | 68.47 | 76.56 | 190.7 |
DeepLabV3 | MobileNetV2 | 71.2 | 81.39 | 5.02 |
PSPNet | ResNet50 | 80.11 | 89.62 | 54.58 |
DeepLabV3 +  | Xception | 80.73 | 89.12 | 60.43 |
DANet | ResNet50 | 78.5 | 87.64 | 48.26 |
SETR | ViT-L | 78.84 | 88.15 | 260.1 |
DAFPN | Swin-T | 80.8 | 89.7 | 158 |
MFAM (Ours) | CA_MobileViT | 80.25 | 88.76 | 9.5 |