cwgan-gp
cwgan-gp copied to clipboard
gradient_penalty get negative value
according to the definition of gradient_penalty,it's impossible that get a negative value.why did I get a positive value?
disc_cost = tf.reduce_mean(disc_fake) - tf.reduce_mean(disc_real) + LAMBDA*gradient_penalty
gradient_penalty can't be negative but the other 2 terms can be.