Fig 1.
The network architecture of our MPAN.
Fig 2.
The structure of the proposed concatenated attention module.
Fig 3.
The structure of the proposed integrated attention module.
Fig 4.
Pseudocode of MPAN in a PyTorch-like style.
Fig 5.
Visual comparison for X3 SR on our face dataset.
The best results are bold. Our network achieves superior performance and recovers more face details than previous networks.
Table 1.
Ablation study about the proposed CAM and IAM.
Table 2.
Ablation study about IRAG numbers.
Table 3.
Ablation study about RCAB numbers.
Fig 6.
Visual comparison for 4X SR on our face dataset.
The best results are bold.
Table 4.
Quantitative results with 128x128 output and scale factor of X2, X3 and X4.
The top and second-place results are emphasized in bold and underlined, respectively.
Fig 7.
Comparative results of MPAN and RDN for 4X SR.
Fig 8.
Comparative results of MPAN and SPARNet for 2X SR.
Fig 9.
Comparative results of MPAN and SAN for 4X SR.