search

TachibanaYoshino / AnimeGANv2

[Open Source]. The improved version of AnimeGAN. Landscape photos/videos to anime
5.06k stars 709 forks source link

when I train my dataset: mean_v_loss: nan #71

Open FanYans opened 5 months ago

FanYans commented 5 months ago

When I train my dataset, the mean_v_loss became nan. How to solve this issue?

TachibanaYoshino commented 4 months ago

I did not have this error during training. You can check the calculation of the loss function and whether the value output by the network is abnormal.

FanYans commented 4 months ago

Thank you very much！ I will try it. However when Ｉ

------------------ 原始邮件 ------------------ 发件人: "TachibanaYoshino/AnimeGANv2" @.>; 发送时间: 2024年4月10日(星期三) 下午4:20 @.>; @.**@.>; 主题: Re: [TachibanaYoshino/AnimeGANv2] when I train my dataset: mean_v_loss: nan (Issue #71)

I did not have this error during training. You can check the calculation of the loss function and whether the value output by the network is abnormal.

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

FanYans commented 4 months ago

Thank you very much！ I will try it. However when finished my training (10 epoch), my results were black pictures. I do not why? Have you ever had this problem before？Maybe I don't train enough？

------------------ 原始邮件 ------------------ 发件人: "TachibanaYoshino/AnimeGANv2" @.>; 发送时间: 2024年4月10日(星期三) 下午4:20 @.>; @.**@.>; 主题: Re: [TachibanaYoshino/AnimeGANv2] when I train my dataset: mean_v_loss: nan (Issue #71)

I did not have this error during training. You can check the calculation of the loss function and whether the value output by the network is abnormal.

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

TachibanaYoshino commented 4 months ago

Something is wrong with your training. The model does not converge correctly. Check your data processing and data entry network for problems. This problem does not occur under normal circumstances.

FanYans commented 4 months ago

I found it's an issue with my dataset. I'll check the training results after I finish training. Thank you very much to the author. My training progress is as shown in the graph below. Is this loss rate trend normal?

------------------ 原始邮件 ------------------ 发件人: "TachibanaYoshino/AnimeGANv2" @.>; 发送时间: 2024年4月11日(星期四) 上午10:48 @.>; @.**@.>; 主题: Re: [TachibanaYoshino/AnimeGANv2] when I train my dataset: mean_v_loss: nan (Issue #71)

Something is wrong with your training. The model does not converge correctly. Check your data processing and data entry network for problems. This problem does not occur under normal circumstances.

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