search

BR-IDL / PaddleViT

:robot: PaddleViT: State-of-the-art Visual Transformer and MLP Models for PaddlePaddle 2.0+
https://github.com/BR-IDL/PaddleViT
Apache License 2.0
1.22k stars 318 forks source link

ViT模型验证时要求数据集必须有训练集和验证集不太合理 #87

Closed xnli closed 2 years ago

xnli commented 2 years ago

Describe the bug 在使用ViT基于Imagenet2012数据集进行模型验证时，因为数据集比较大，仅下载了验证数据集。执行模型验证命令，报错，提示没有训练集数据。

经排查， main_single_gpu.py 在进行模型验证前，需要先加载训练集和验证集。

建议单独增加一个模型验证的脚本。

xperzy commented 2 years ago

Thanks for the issue!
Yes, that is a real problem if someone only needs to run the validation without touching the training data.

I will fix this and update the code as soon as possible!

lmk123568 commented 2 years ago

Describe the bug 在使用ViT基于Imagenet2012数据集进行模型验证时，因为数据集比较大，仅下载了验证数据集。执行模型验证命令，报错，提示没有训练集数据。

经排查， main_single_gpu.py 在进行模型验证前，需要先加载训练集和验证集。

建议单独增加一个模型验证的脚本。

你的建议非常重要

xperzy commented 2 years ago

@xnli This issue has been fixed in ViT, will update to other models asap. Please see commit 9a7c105 for detailed changes.

I will start a new issue to update other main scripts, so I close this issue