Breaking the Dilemma of Medical Image-to-image Translation Lingke Kong

All the experiments were implemented in Pytorch software on 64-bit Ubuntu Linux system with 96GB RAM and 24GB Nvidia Titan RTX GPU. All the images were normalized to [-1, 1] and resampled to 256 256. The batch size was set to 1 with weight decay 1e-4. The training process includes totally 80 epochs and over 640K iterations. We also set different weights for different loss functions, as shown in Table 1 .