Helping regression task with a discriminator

Hello everyone. I would like to add a discriminator as an extra loss, in order to enhance my current network that is trained with a regression task. Currently i have 4 losses(2 L1 smooth, 1 MSE and one SSIM). Though, until now I haven’t seen any progress. Has anyone such an experience or could help me with any idea? I use weight=0.05 for the discriminator loss.