Training with CIFAR-10

[hosytuyen]

Thank you for the great work.

Besides, Does anyone try to train with CIFAR-10. I have followed the setup for Mnist: replace the data loader of Mnist to CIFAR-10, change input dimension from 1 to 3, keep the same models. However, the result is not good (about 31%) on FedAvg.

Is there any special setting when do experiment with a new dataset? Thank you

[zhuangdizhu]

Hi, for this project we did not implement on the cifar10. I suggest using larger training networks say resNet 18 for the cifar10 dataset, and accordingly, the generator may need to be extended with more layers to capture such domain knowledge.

[gaurav0190]

Thank you for the great work.

Besides, Does anyone try to train with CIFAR-10. I have followed the setup for Mnist: replace the data loader of Mnist to CIFAR-10, change input dimension from 1 to 3, keep the same models. However, the result is not good (about 31%) on FedAvg.

Is there any special setting when do experiment with a new dataset? Thank you

can you pl share in which file you change for cifar. i have done in model_utils , model_config and run the code But there is an error. return torch._C._nn.linear(input, weight, bias) RuntimeError: mat1 and mat2 shapes cannot be multiplied (300x1024 and 784x32) where should i change it for cifar i keep setting same as mnist used with channel 1 to 3

[LonelyMoonDesert]

same problem with you...I also got 31% accuracy on CIFAR10 with resnet56

[alonez12]

Now I have a question.About cifar - 10 ，I'm going to plug in cifar-10 and modified the file（model_utils.py）,but There are many problems and it do not work.So can you see how the code changes in cifar? Thank you very much
RuntimeError: mat1 and mat2 shapes cannot be multiplied (300x128 and 2048x64)

[LonelyMoonDesert]

你有微信或者QQ嘛，感觉留个联系方式聊起来更方便

------------------ 原始邮件 ------------------ 发件人: "zhuangdizhu/FedGen" @.>; 发送时间: 2022年5月9日(星期一) 下午5:18 @.>; @.**@.>; 主题: Re: [zhuangdizhu/FedGen] Training with CIFAR-10 (Issue #5)

抱歉 打扰了 请问下cifar10是如何修改的呢 我修改之后发现RuntimeError: mat1 and mat2 shapes cannot be multiplied (300x128 and 2048x64)

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

[alonez12]

我的qq为35674047 十分感谢

------------------ 原始邮件 ------------------ 发件人: "zhuangdizhu/FedGen" @.>; 发送时间: 2022年5月9日(星期一) 晚上11:31 @.>; @.**@.>; 主题: Re: [zhuangdizhu/FedGen] Training with CIFAR-10 (Issue #5)

你有微信或者QQ嘛，感觉留个联系方式聊起来更方便

------------------ 原始邮件 ------------------ 发件人: "zhuangdizhu/FedGen" @.>; 发送时间: 2022年5月9日(星期一) 下午5:18 @.>; @.**@.>; 主题: Re: [zhuangdizhu/FedGen] Training with CIFAR-10 (Issue #5)

抱歉 打扰了 请问下cifar10是如何修改的呢 我修改之后发现RuntimeError: mat1 and mat2 shapes cannot be multiplied (300x128 and 2048x64)

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

[gaurav0190]

@Crestiny Crestiny... any one who have solved the problem.

[CityChan]

same problem with you...I also got 31% accuracy on CIFAR10 with resnet56

Could you share the code for cifar10 with resnet18?

[CityChan]

Now I have a question.About cifar - 10 ，I'm going to plug in cifar-10 and modified the file（model_utils.py）,but There are many problems and it do not work.So can you see how the code changes in cifar? Thank you very much RuntimeError: mat1 and mat2 shapes cannot be multiplied (300x128 and 2048x64)

same problem, did you solve the problem?

[LonelyMoonDesert]

@Crestiny Crestiny... any one who have solved the problem.

sorry for reply late but as far as i know, no one :(

[LonelyMoonDesert]

same problem with you...I also got 31% accuracy on CIFAR10 with resnet56

Could you share the code for cifar10 with resnet18?

你有微信或者QQ嘛，或许可以留个联系方式呀www

[Xinchungeorb]

@Crestiny Crestiny... any one who have solved the problem.

sorry for reply late but as far as i know, no one :(

请问一下在cifar10上使用其他模型的准确率是不是也很低呢？

[LonelyMoonDesert]

@Crestiny Crestiny... any one who have solved the problem.

sorry for reply late but as far as i know, no one :(

请问一下在cifar10上使用其他模型的准确率是不是也很低呢？

我只测试过resnet56这样子，不过我想resnet深度是9n+2这个系列的应该都是同样精度不高。其他架构就没有试过啦。

[CityChan]

@Crestiny Crestiny... any one who have solved the problem.

sorry for reply late but as far as i know, no one :(

请问一下在cifar10上使用其他模型的准确率是不是也很低呢？

我只测试过resnet56这样子，不过我想resnet深度是9n+2这个系列的应该都是同样精度不高。其他架构就没有试过啦。

在cifar10上跑resnet5的代码能不能share?

[LonelyMoonDesert]

@Crestiny Crestiny... any one who have solved the problem.

sorry for reply late but as far as i know, no one :(

请问一下在cifar10上使用其他模型的准确率是不是也很低呢？

我只测试过resnet56这样子，不过我想resnet深度是9n+2这个系列的应该都是同样精度不高。其他架构就没有试过啦。

在cifar10上跑resnet5的代码能不能share?

实在是有点混乱……不太拿得出手哈哈哈，阅读起来可能很困扰

[MeteorBless]

tion.About cifar - 10 ，I'm going to plug in cifar-10 and modified the file（model_utils.py）

可以分享一下代码吗？这是我的邮箱2677017641@qq.com。非常感谢！