Hey,
I've been trying to use the hdf5 format with caffe.
Just to check, if I have N images which are in colour and WxH pixels in size, and each image has a 1xL label, I should follow the following steps:
Create a numpy array which is (N,W,H,3) in shape.
Instantiate each of the N positions with an image.
Create a numpy array which is (N,L) in shape.
Instantiate each of the N positions with L labels.
Save these arrays in a .h5 file as 'data' and 'label' respectively.
Train away.
Could someone tell me if there is an issue with this process. The documentation is a 'touch' unclear.
Thanks!