Training 200 rounds without convergence

[zyxzyx45]

The self-dataset used is optical image and SAR image, the parameter is selected u_net, the image size parameter and cropping size are changed, and the rest of the loss parameters are not changed, and the training is L1_TR to the end, the GAN_TR, the L1_RT, and the GAN_RT are all about 20, and the rest are all 0, why? The results of this training test are very poor, and they are all gray images with no texture at all.