search

leftthomas / ImageDeblurring

A Keras implementation of image deblurring based on ICCV 2017 paper "Deep Generative Filter for motion deblurring"
82 stars 30 forks source link

why the d_loss tends to be zero and how to express the adversarial loss as the paper said #30

Closed Alukassh closed 5 years ago

Alukassh commented 5 years ago

Sorry to bother you. When trained for a few epoch, the d_loss tends to be zero, does that means the generator has been trained well enough or there are something wrong with the discriminator loss expressions? Another question is, how to express the adversarial loss as the paper said? someone has any ideas ？

leftthomas commented 5 years ago

@Alukassh This is not the problem about the code, you could discuss these with others on reddit