How to train supernet on my own dataset?

[JiuTMniJiaoXiaLuo]

Checklist

• I have searched related issues but cannot get the expected help.
• I have read related documents and don't know what to do.

Describe the question you meet

Most of the existing datasets that can train supernet are ImageNet and CIFAR.What configs files should be modified in order to replace them with other datasets?

Post related information

[pprp]

Hi, @JiuTMniJiaoXiaLuo , thank you for your interests.

As for your question, please refer to the mmcls document.

[JiuTMniJiaoXiaLuo]

Hi, @JiuTMniJiaoXiaLuo , thank you for your interests.

As for your question, please refer to the mmcls document.

Thank you for your reply,but after I created my dataset according to the steps, a new problem occurred.

[pprp]

Hi, @JiuTMniJiaoXiaLuo

You can refer to the similar issue: https://github.com/open-mmlab/mmtracking/issues/539

[JiuTMniJiaoXiaLuo]

Hi, @JiuTMniJiaoXiaLuo

You can refer to the similar issue: open-mmlab/mmtracking#539 Thank you for helping me, but I didn't understand. Which parameter did they delete and where did they delete it ?

[pprp]

Hi, @JiuTMniJiaoXiaLuo , I think they delete the file_client_args parameter.

Maybe the train_pipeline in your config is like this:

train_pipeline = [
    dict(type='LoadImageFromFile', file_client_args=file_client_args),
    ...
]

You can change to:

train_pipeline = [
    dict(type='LoadImageFromFile'),
    ...
]

[JiuTMniJiaoXiaLuo]

Hi, @JiuTMniJiaoXiaLuo , I think they delete the file_client_args parameter.

Maybe the train_pipeline in your config is like this:

train_pipeline = [
    dict(type='LoadImageFromFile', file_client_args=file_client_args),
    ...
]

You can change to:

train_pipeline = [
    dict(type='LoadImageFromFile'),
    ...
]

Thank you very much!