1

I am fine tuning a neural net that takes 2 input images, which are later concatenated. To fine tune, I would like to perform regression and add labels to this network.

How can I do this? Currently, the net uses an ImageData layer for input which I will have to change to HDF5. But how can I pass a single label to these two input channels?

I have looked around at how to do this and I found this question. However, here they create a single HDF5 file with 2 input images. In my case, I'd have two HDF5 files with 1 input image, which are then concatenated and then has 1 to the same label.

Community
  • 1
  • 1
Cassie
  • 362
  • 4
  • 15
  • 1
    you can put everything in a single HDF5 file with multiple datasets – Shai Jun 16 '16 at 13:27
  • Thanks for the quick reply! I have one more question: if I have `{"data1":[images], "data2":[images]}`, how can I pass labels? – Cassie Jun 16 '16 at 13:47
  • 1
    why not `{"data1":[images], "data2":[images], "label":[labels]}`? You are not restricted to two datasets in a single HDF5 file... – Shai Jun 16 '16 at 14:37

0 Answers0