How to create dataset for image regression training

Hi all,

I would like to train a deep learning network on DIGITS but not sure how to prepare the data. The network takes a RGB image as input and gives two floating pointer number as output. I have collected training and validation samples and organized them in the following directories:

----data
  ----train
    ----images
        00000.jpg
        00001.jpg
        00002.jpg
        ...
    ----labels
        00000.txt
        00001.txt
        00002.txt
        ...
  ----val
    ----images
        00000.jpg
        00001.jpg
        00002.jpg
        ...
    ----labels
        00000.txt
        00001.txt
        00002.txt
        ...

label file xxxxx.txt are text file with two numbers in it, for example
0.5678, 1.2345

Could anyone tell me how to go from here and create a dataset for DIGITS?

thanks,