U-Net Fast-Forward Operator
The fully convolutional design allows the model to process arbitrary spatial input sizes. The U-Net represents complex object fields as two real-valued channels for real and imaginary components. The model is trained using early-stage iter…
1 sources - 4 claims
The fully convolutional design allows the model to process arbitrary spatial input sizes. The U-Net represents complex object fields as two real-valued channels for real and imaginary components. The model is trained using early-stage iteration-5 reconstructions as inputs and later iteration-100 reconstructions as targets. The fast-forward model is a U-Net implemented with the dlsia library.