1

I am very new to using the caffe framework. I have a network which works on images. It takes images as input and outputs images. Does caffe have a layer I have to include so as to save the output images?

Thank you.

user3126802
  • 419
  • 8
  • 19

1 Answers1

1

Caffe doesn't have a "save-to-image" layer.
You can try and see if "HDF5Output" layer can work for you: you can save your output image as a floating-point ND array into hdf5 binary file and read it later with other applications.

Alternatively, you can use a "Python" layer to save the images for you. Just write the save code in python (using PIL.Image package for example).

Shai
  • 111,146
  • 38
  • 238
  • 371
  • Thank you for your reply. Would it make sense to try and read the blob data from a deconvolution layer and save that? – user3126802 Dec 27 '17 at 12:02
  • @user3126802 I don't know what your are trying to do. As long as it makes sense to you - you are welcome to try and do it. – Shai Dec 27 '17 at 12:05
  • Thank you for your help! – user3126802 Dec 27 '17 at 12:06
  • I was reading up about the HDFOutput layer. However a problem may arise, since it would try to create the same file for each iteration. – user3126802 Dec 27 '17 at 13:19
  • @user3126802 I'm afraid I do not have any experience with `"HDF5Output"` layer, I did hear about issues people had using this layer. I suspect it would be easier to use a `"Python"` layer for this task. – Shai Dec 27 '17 at 13:20