.mat files as input for caffe deep learning network

Question

Is it possible to use .mat files to store input data for the caffe deep learning framework? if yes, how can I change the 'data' of my train_val.prototxt? I am using the googlenet network.

If no, which other file formats can I use rather than mdb and hd5?

For me, using .mat files is better because of their size, my .hd5 files are huge (caffe can't read them because of lack of memory) and I can't find how to save my data as .mdb files in Matlab.

Answer 1

Mat files version 7.3 are hdf5 files, just make sure to use this format when writing. (check documenting for save)

Any hdf5 library which supports gzip compression can read mat files.