search

yuangan / EAT_code

Official code for ICCV 2023 paper: "Efficient Emotional Adaptation for Audio-Driven Talking-Head Generation".
Other
275 stars 31 forks source link

The training code #7

Closed julia-cherry closed 10 months ago

julia-cherry commented 1 year ago

When will you release the training code? Thanks very much!

yuangan commented 1 year ago

Thank you for your attention. We have been busy recently. The training code may be released in the next month.

yuangan commented 1 year ago

We have released the Emotional Adaptation Training code. Thank you for your patience~

YZX-codesky commented 12 months ago

When will the complete training code containing two stages be released? Thank you very much！

yuangan commented 11 months ago

Thank you for your attention. We will release the A2ET training code next month, which will require several days to train with VoxCeleb2.

yuangan commented 10 months ago

We have now released all of the training code. If you have any questions about the training, please feel free to ask me.

YZX-codesky commented 10 months ago

您好，作者，非常感谢您的工作！我在训练第二阶段的代码的时候，发现训练的时间变得非常长，长达800多个小时，见下图：

我采用的是一张3090显卡，batch_size也改为了1。想问一下这个是正常的吗？

------------------ 原始邮件 ------------------ 发件人: "yuangan/EAT_code" @.>; 发送时间: 2023年12月26日(星期二) 晚上10:59 @.>; @.**@.>; 主题: Re: [yuangan/EAT_code] The training code (Issue #7)

Closed #7 as completed.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

yuangan commented 10 months ago

我们用了4张3090卡训练，按半个小时存一个epoch，一般来说第二阶段训练5-8个小时就行，具体训练时间和performance的关系可以参考我们论文中的Fig.5。1张3090可能需要训久一点，你可以根据训练loss和其他指标来判断需要训练多久。