/AFPN

Adaptive Detail Injection-Based Feature Pyramid Network For Pan-sharpening

Primary LanguagePython

AFPN (updating...)

Adaptive Detail Injection-Based Feature Pyramid Network For Pan-sharpening

Fig.1 : the overall framework of AFPN.

image

Fig.1 caption: Green and orange layers are spectral and spatial feature maps of spectral and spatial pyramids respectively. Resblock layer contains two convolutional layers, a PReLu activation function, and a residual connection. Upsample layer is Bicubic interpolation and Resample layer is a convolutional layer with stride=2.

TI vs. FDI on AFPN

image

More experiments

Texture loss

image

Inference Time and Params

image

ADI on BPN

image

Mistakes in Paper

image