What is the correct value?

[taki0112]

In your paper, you use a 800 lambda for the image reconstruction loss. but, in readme, you use a 800 lambda for the latent reconstruction loss.

where should I use 800 ? image_rec ? or latent_rec ?

[yaxingwang]

Hi @taki0112 , Thanks. There is a type error. The parametric for latent_rec (noise) is configured to 800, while the image_rec is 10. I will update the project to mention this error

[yaxingwang]

I have update it.

[taki0112]

@yaxingwang I saw your update. Thank you

one more.

1. I think you should remove #(https://github.com/yaxingwang/SDIT/blob/master/solver.py#L273). What do you think ?
2. And in your paper, you use 10 lambda for adversarial_loss. but in code, you use 1. I think 1 is correct. right?

[yaxingwang]

Thanks. I will update it.

[taki0112]

if you don't use a noise_loss for the discriminator, then you also remove this. d_loss_noise

[yaxingwang]

In fact we try different configures to d_loss_noise, and we use it. I will finally check this configure.