Determining loss values during training

chrisdonahue / wavegan

WaveGAN: Learn to synthesize raw audio with generative adversarial networks

MIT License

1.32k stars 283 forks source link

Determining loss values during training #35

Closed karanpareek96 closed 5 years ago

karanpareek96 commented 5 years ago

Hi Chris,

How can I determine the loss values for the Generator and DIscriminator during the training process?

chrisdonahue commented 5 years ago

These values are displayed on the "Scalars" tab of tensorboard as G_loss and D_loss. Their meaning changes depending on which type of GAN you are training.

saraalemadi commented 5 years ago

Hi @chrisdonahue ,

I am getting negative values in the Scalar tab of TensorBoard for G_Loss and D_Loss, what does that indicates?

chrisdonahue commented 5 years ago

That is normal. They reflect the discriminator's assessment of the 1-Wasserstein distance between the real and the generated probability distributions. I haven't been able to correlate these values to any observable behavior of the waveforms during training TBH.