Open Guanys-dar opened 3 weeks ago
JianzeLi-114
commented
2 weeks ago Thank you very much for your interest in our work!
The latent vector input to the generator is the result of the RGB image encoded by the VAE, without adding any random noise. Randomly sampled timesteps only occur in the discriminator, which does not appear during the inference process. For both the training and inference processes, the generator uses a fixed timestep for denoising.
Guanys-dar
commented
2 weeks ago Thank you very much for your clear response, it has been incredibly helpful to me. Thank you!
2024-10-22 09:28:23 "JianzeLi" @.***> 写道：
Thank you very much for your interest in our work!
The latent vector input to the generator is the result of the RGB image encoded by the VAE, without adding any random noise. Randomly sampled timesteps only occur in the discriminator, which does not appear during the inference process. For both the training and inference processes, the generator uses a fixed timestep for denoising.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
Thank you for your excellent and inspiring work!
I have a question: during training, is the noise level applied to the latent input of the generator randomly sampled between 1 and 1000? How is the noise level handled during inference? Is it also randomly sampled, or is there a fixed value used?
Looking forward to your response, and thank you in advance!