Does anyone have ImageNet data peparation script?

[Ilya Zhenin]

I want to pretrain my model on ImageNet classification task, and then use this model for other tasks. I know there is pretrained models in Caffe Zoo, but I want to make my own.

But decent classification pretraining on ImageNet is a huge task itself, and I do not have time to invest it in. I mean there is a lot of time for figuring out which image resolution will be best, mean substraction, most tedious and not trivial - data augmentation(rotattion, mirroring, shifting, etc).

So I wonder if you asked this question to yourself - how to create suitable for most networks architectures Image Net data preparation, which you can feed into network training and get good results and proceed for your main task.

Methodic or script will be great.

Saying it using concreate example - how to train ResNet or VGG-16 from zero point and achieve same results as authours?