Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

The retrained results have color differences and unremovable mask shadows. #145

Open
xuanandsix opened this issue Sep 2, 2024 · 15 comments

Comments

@xuanandsix
Copy link

WeChat73ba3211726c21a91433398b4f3885bb

some suggestions, please.

@all1new
Copy link

all1new commented Sep 4, 2024

How many cards did you use for training? Will NCCL timeout occur?

@xuanandsix
Copy link
Author

How many cards did you use for training? Will NCCL timeout occur?

6 * A800, NCCL is work. More training steps improve the color and mask problem.

@all1new
Copy link

all1new commented Sep 5, 2024

What is the final effect? ​​Will it be better than the original weight of IDM-VTON?

@awais-nayyar
Copy link

I want to train this code on my own dataset. What is the minimum GPU and Ram requirements???
can anybody help to guide me?

@xuanandsix
Copy link
Author

What is the final effect? ​​Will it be better than the original weight of IDM-VTON?

For some pictures, there is still color difference. The color of the clothes is not as bright as the image prompt.

WeChat7147d970e0627b928e0b22fad41261c8

@all1new
Copy link

all1new commented Sep 5, 2024

What is the final effect? ​​Will it be better than the original weight of IDM-VTON?

For some pictures, there is still color difference. The color of the clothes is not as bright as the image prompt.

WeChat7147d970e0627b928e0b22fad41261c8

The conclusion of my training is the same as yours. Is there any way to improve it?

@TnoobT
Copy link

TnoobT commented Sep 14, 2024

What is the final effect? ​​Will it be better than the original weight of IDM-VTON?

For some pictures, there is still color difference. The color of the clothes is not as bright as the image prompt.
WeChat7147d970e0627b928e0b22fad41261c8

The conclusion of my training is the same as yours. Is there any way to improve it?

问题解决了吗,我训练了一天还是这样,loss感觉也没收敛

@jiayoujiayoujiayoua
Copy link

I have encountered the same problem. Has anyone solved this problem? Can you give me some advice? I would be very grateful.

@TnoobT
Copy link

TnoobT commented Sep 25, 2024

I have encountered the same problem. Has anyone solved this problem? Can you give me some advice? I would be very grateful.

  1. 不要用8bit adam优化器;
  2. 用他的训练代码;https://github.com/luxiaolili/IDM-VTON-train

@jiayoujiayoujiayoua
Copy link

I have encountered the same problem. Has anyone solved this problem? Can you give me some advice? I would be very grateful.

  1. 不要用8bit adam优化器;
  2. 用他的训练代码;https://github.com/luxiaolili/IDM-VTON-train
    非常感谢你,我马上再训练一下试试,好人一生平安!

@all1new
Copy link

all1new commented Sep 25, 2024

I have encountered the same problem. Has anyone solved this problem? Can you give me some advice? I would be very grateful.

  1. 不要用8bit adam优化器;
  2. 用他的训练代码;https://github.com/luxiaolili/IDM-VTON-train
    非常感谢你,我马上再训练一下试试,好人一生平安!

我用这个尝试过, 速度有些慢, 从基础模型开始训练试一下, 我这边还是没法复现模型效果

@jiayoujiayoujiayoua
Copy link

jiayoujiayoujiayoua commented Sep 25, 2024 via email

@TnoobT
Copy link

TnoobT commented Sep 25, 2024

I have encountered the same problem. Has anyone solved this problem? Can you give me some advice? I would be very grateful.

  1. 不要用8bit adam优化器;
  2. 用他的训练代码;https://github.com/luxiaolili/IDM-VTON-train
    非常感谢你,我马上再训练一下试试,好人一生平安!

我用这个尝试过, 速度有些慢, 从基础模型开始训练试一下, 我这边还是没法复现模型效果

还好吧,训5000个step就能inpainting了,整张图的色差也基本没有了

@all1new
Copy link

all1new commented Sep 25, 2024

I have encountered the same problem. Has anyone solved this problem? Can you give me some advice? I would be very grateful.

  1. 不要用8bit adam优化器;
  2. 用他的训练代码;https://github.com/luxiaolili/IDM-VTON-train
    非常感谢你,我马上再训练一下试试,好人一生平安!

我用这个尝试过, 速度有些慢, 从基础模型开始训练试一下, 我这边还是没法复现模型效果

还好吧,训5000个step就能inpainting了,整张图的色差也基本没有了

有没有对比过和IDM-VTON原始权重的效果, 以及对各种款式比如"毛衣", "图案复杂服饰", 线上模特; 这样多维度对比效果;
如果有的话 效果是否一致呢?

@all1new
Copy link

all1new commented Sep 25, 2024

好的佬,我主要是在自己的数据集上面做微调 不知道效果咋样 佬 加个联系方式吗 我训练有结果了一起讨论讨论

------------------ Original ------------------ From: 1_corn @.> Date: Wed,Sep 25,2024 4:54 PM To: yisol/IDM-VTON @.> Cc: jiayoujiayoujiayoua @.>, Comment @.> Subject: Re: [yisol/IDM-VTON] The retrained results have color differences andunremovable mask shadows. (Issue #145) I have encountered the same problem. Has anyone solved this problem? Can you give me some advice? I would be very grateful. 不要用8bit adam优化器; 用他的训练代码;https://github.com/luxiaolili/IDM-VTON-train 非常感谢你,我马上再训练一下试试,好人一生平安! 我用这个尝试过, 速度有些慢, 从基础模型开始训练试一下, 我这边还是没法复现模型效果 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

嗯嗯 加 我邮箱是[email protected]

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants