自己训练stage2，stage3模型，推理时，设置w不等于0和1时，嘴唇会出现色块，有点偏绿色，紫色等

[create-li]

请问，基于vqgan训练stage3后，推理时，w不设置0,1时，就会出现人嘴唇的出现色块，但是w=0,1时就没有问题。

[smartaji]

Tamil please

On Tue, Feb 27, 2024, 12:09 PM create-li @.***> wrote:

请问，基于vqgan训练stage3后，推理时，w不设置0,1时，就会出现人嘴唇的出现色块，但是w=0,1时就没有问题。

— Reply to this email directly, view it on GitHub https://github.com/sczhou/CodeFormer/issues/354, or unsubscribe https://github.com/notifications/unsubscribe-auth/A5T6PMQZGYDB7O6LI2FOWS3YVV5QHAVCNFSM6AAAAABD3QOZMKVHI2DSMVQWIX3LMV43ASLTON2WKOZSGE2TKNZYGY3DQNI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[FantasticZihao]

Hello，I have some quetions about how to train the codeformer on my datasets. I know the first stage is to train the vqgan, so after i have done this, must i use the data which used in the first stage to generate latent gt and use the latent gt to train the next stage?

[FantasticZihao]

@create-li I am deperated to your help. Can you have my QQ:3560614254?