data augmentation about caffe-vdsr HOT 4 CLOSED

I trained my "VDSR_170000" model by about 300000 training samples (1638 images augumented from 91 images).
You can easily get 1638 images from 91 images by simple transforms, like rotation, crop, flip and translation。

from caffe-vdsr.

thanks for your quick reply, i understand now. have you ever try real time data augmentation in Caffe? do you think it is better than preprocessing the traing data? similar with this https://github.com/kevinlin311tw/caffe-augmentation

from caffe-vdsr.

I didn't try real time data augmentation in caffe. You can have a try. In my opinion, it wil be more effective.

from caffe-vdsr.

ok i will try to do it. one again thanks

from caffe-vdsr.