Hardness training Imagenet 2012 with VGG16

It’s first time I’m training imagenet dataset with VGG16.

As I know, training dataset that I can download from image-net.org is for both classification and localization.

I want to do is training VGG16 only for classification., but I don’t think this training dataset is appropriate for training classification.

However, I saw several papers that researchers used ILSVRC2012 dataset for classification training.

Also, trying to train usining ILSVRC2012 is failed(low acc) with pytorch official github code.

What is the problem?