Thank you for your excellent work!
During training DCAE using non-image data (specifically data from other physical fields), I observed numerical instability in the latent space. This instability has made it impossible to subsequently train the diffusion model within the latent space. Perhaps it is still necessary to apply a certain degree of regularization to the latent space?
Thank you for your excellent work!
During training DCAE using non-image data (specifically data from other physical fields), I observed numerical instability in the latent space. This instability has made it impossible to subsequently train the diffusion model within the latent space. Perhaps it is still necessary to apply a certain degree of regularization to the latent space?