about sr_trainging and denoising mean_type

[jjjjjgggj]

作者你好，我的数据集在经过dataset处理后，都是(512,512,1)的数据集，然后在sample_trainging中，采用的mean_type为epsilon，然后在随后的denoising中，采用的mean_type也是为epsilon。然后能够正常生成lr_dn的图像。然后在进行sr_training中，第一次采用的mean_type为x0_and_epsilon,denoising中采用的mean_type为epsilon；第二次采用的mean_type为epsilon,denoising中采用的mean_type为epsilon。都出现了RuntimeError: Error(s) in loading state_dict for GaussianDiffusion:。显示的都是模型中的参数与实际的不符合。我想咨询一下作者在训练的时候，采用的是什么值

[DeepXuan]

Based on the information you've provided, I think it's the mean_type difference between training and testing. Can you please provide your detailed configuration error information？

[jjjjjgggj]

你好！这个是报错的一部分内容，大概就是训练的模型参数出现了问题

------------------ 原始邮件 ------------------ 发件人: "DeepXuan/Dn-Dp" @.>; 发送时间: 2024年1月19日(星期五) 晚上11:08 @.>; @.**@.>; 主题: Re: [DeepXuan/Dn-Dp] about sr_trainging and denoising mean_type (Issue #4)

Based on the information you've provided, I think it's the mean_type difference between training and testing. Can you please provide your detailed configuration error information？

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

[jjjjjgggj]

[DeepXuan]

I think the key point is to ensure the training and test configuration files have exactly the same parameters for the network architecture.

[jjjjjgggj]

我认为关键点是确保训练和测试配置文件具有完全相同的网络架构参数。

你的意思是在进行超分部分的时候，训练和测试的channel_multiplier要一致吗？

[DeepXuan]

Absolutely!

[ZhaKiNg]

兄弟 请问你知道sr3_moudle和ddpm_moudle有什么区别吗 整个打包的文件里面好像缺失了ddpm模块的内容

[ZhaKiNg]

兄弟 请问你知道sr3_moudle和ddpm_moudle有什么区别吗 整个打包的文件里面好像缺失了ddpm模块的内容

还有一个问题就是使用自己的数据集进行训练的时候yaml文件中是只需要改一个模型吗？

[zhizi20]

channel_multiplier

请问你用预训练模型进行测试了吗？有没有结果

[jjjjjgggj]

我训练不出来…参数一直不对。后来换方向了..就没有看这个了

发自我的iPhone

------------------ Original ------------------ From: zhizi20 @.> Date: Sat,Jul 13,2024 10:33 PM To: DeepXuan/Dn-Dp @.> Cc: jjjjjgggj @.>, Author @.> Subject: Re: [DeepXuan/Dn-Dp] about sr_trainging and denoising mean_type(Issue #4)

channel_multiplier

请问你用预训练模型进行测试了吗？有没有结果

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

[zhizi20]

我训练不出来…参数一直不对。后来换方向了..就没有看这个了 发自我的iPhone … ------------------ Original ------------------ From: zhizi20 @.> Date: Sat,Jul 13,2024 10:33 PM To: DeepXuan/Dn-Dp @.> Cc: jjjjjgggj @.>, Author @.> Subject: Re: [DeepXuan/Dn-Dp] about sr_trainging and denoising mean_type(Issue #4) channel_multiplier 请问你用预训练模型进行测试了吗？有没有结果 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

我也没有训练出来，后来我用预训练的模型去测试，也是一直报错预训练模型架构与当前模型架构不一致，不知道怎么修改当前模型架构。修改了配置文件还是没有解决。你现在换了什么方向？

[jjjjjgggj]

他有两部分模型，第一部分我是没有问题的，第二部分一直不行。现在做老师的课题去了

发自我的iPhone

------------------ Original ------------------ From: zhizi20 @.> Date: Sat,Jul 13,2024 10:38 PM To: DeepXuan/Dn-Dp @.> Cc: jjjjjgggj @.>, Author @.> Subject: Re: [DeepXuan/Dn-Dp] about sr_trainging and denoising mean_type(Issue #4)

我训练不出来…参数一直不对。后来换方向了..就没有看这个了 发自我的iPhone … ------------------ Original ------------------ From: zhizi20 @.> Date: Sat,Jul 13,2024 10:33 PM To: DeepXuan/Dn-Dp @.> Cc: jjjjjgggj @.>, Author @.> Subject: Re: [DeepXuan/Dn-Dp] about sr_trainging and denoising mean_type(Issue #4) channel_multiplier 请问你用预训练模型进行测试了吗？有没有结果 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

我也没有训练出来，后来我用预训练的模型去测试，也是一直报错预训练模型架构与当前模型架构不一致，不知道怎么修改当前模型架构。修改了配置文件还是没有解决。你现在换了什么方向？

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

[huster2022]

@jjjjjgggj 请问你第一部分怎么设置的训练参数呀，我跑自己的数据集，训练过程中指标不怎么变化，学习率增大减小都试过