Loading Street View House Numbers Dataset in Python

Question

I am trying to work with the Street View House Numbers Dataset in Python. To load the .mat file digitStruct.mat in python I am using H5py

import h5py
labels_file = './sv/train/digitStruct.mat'
f = h5py.File(labels_file)
struct= f.values()
names = struct[1].values()
print(names[1][1].value)

I get [<HDF5 object reference>] but I need to know the actual string

Answer 1

If you already have a <HDF5 object reference>, and you want to know the real data, here is one approach:

import h5py
labels_file = './sv/train/digitStruct.mat'
f = h5py.File(labels_file)
ref = f.get('/digitStruct/name').value[0][0]  # <HDF5 object reference>
print(f[ref].value)

In my environment, f[ref].value shows:

In [84]: f[ref].value
Out[84]: 
array([[ 49],
       [ 46],
       [112],
       [110],
       [103]], dtype=uint16)

For more information on object references, you can refer to http://docs.h5py.org/en/latest/refs.html

Answer 2

I have not checked the dataset but this is what I suggest. Have a look at this method - get to extract.

Apart from this, I also suggest checking Matlab engine for python.

eng = matlab.engine.start_matlab()
data = eng.load(labels_file,nargout=1)