clearlyzero comments

Results 33 comments of


                                            clearlyzero

How generate image from noise vector with KL-reg autoencoder

> ![image](https://github.com/CompVis/latent-diffusion/assets/65875405/b499b7d1-2471-405e-90c7-50a2fc13cce7) like this

Proper Learning Rate decay for training the LDM

May I ask whether you encountered this problem in the first stage or the second stage?

Proper Learning Rate decay for training the LDM

Me too, I also encountered this problem. I trained many times but the loss still remained at 0.4.Have you solved this problem?

Proper Learning Rate decay for training the LDM

Is the initial learning rate set to 5e-5?

Proper Learning Rate decay for training the LDM

Thank you. Thank you very much for your reply. I will try it right away.

Proper Learning Rate decay for training the LDM

I tried it and found that it still didn't work. The loss value still remained at 0.4.

Proper Learning Rate decay for training the LDM

I feel like there is something wrong with my X0 processing x_start = AutoEncoderKl.Encoder(x_start).sample() ``` def p_losses(self, x_start, t,img_r, noise = None): b, c, h, w = x_start.shape noise =...

multi-gpus with max_batch_size > 1 with ddp

I still have this problem. Have you solved it?

Pre-Train Model

Hello, I'm sorry to bother you. My training process discriminator loss will remain at 2. Did you encounter this situation during the training of this network.

Pre-Train Model

> > Its been awhile since I did the training, but if I remember correctly, yes, somewhat? It should still bounce around, but the loss stays relatively high. Thanks,Does this...