import numpy as np
import scipy.io as sio
# Taken from http://stackoverflow.com/questions/7008608/scipy-io-loadmat-nested-structures-i-e-dictionaries
def loadmat(filename):
'''
this function should be called instead of direct sio.loadmat
as it cures the problem of not properly recovering python dictionaries
from mat files. It calls the function check keys to cure all entries
which are still mat-objects
'''
data = sio.loadmat(filename, struct_as_record=False, squeeze_me=True)
return _check_keys(data)
def _check_keys(dict):
'''
checks if entries in dictionary are mat-objects. If yes
todict is called to change them to nested dictionaries
'''
for key in dict:
if isinstance(dict[key], sio.matlab.mio5_params.mat_struct):
dict[key] = _todict(dict[key])
return dict
def _todict(matobj):
'''
A recursive function which constructs from matobjects nested dictionaries
'''
dict = {}
for strg in matobj._fieldnames:
elem = matobj.__dict__[strg]
if isinstance(elem, sio.matlab.mio5_params.mat_struct):
dict[strg] = _todict(elem)
else:
dict[strg] = elem
return dict
if __name__ == '__main__':
model_file = 'LOCATION OF dictnet.mat from http://www.robots.ox.ac.uk/~vgg/research/text/#sec-models NIPS
DLW 2014 models'
mat_contents = loadmat(model_file)
L1 = _todict(mat_contents['layers'][0])
conv1b = np.array(L1['biases'], dtype=np.float32)
conv1W = np.array(L1['filters'], dtype=np.float32)
conv1W = np.reshape(conv1W,(5,5,1,64))
conv1W = np.array(conv1W, dtype=np.float32).transpose((3,2,0,1))[:,:,::-1,::-1]
L1 = _todict(mat_contents['layers'][3])
conv2b = np.array(L1['biases'], dtype=np.float32)
conv2W = np.array(L1['filters'], dtype=np.float32).transpose((3,2,0,1))[:,:,::-1,::-1]
L1 = _todict(mat_contents['layers'][6])
conv3b = np.array(L1['biases'], dtype=np.float32)
conv3W = np.array(L1['filters'], dtype=np.float32).transpose((3,2,0,1))[:,:,::-1,::-1]
L1 = _todict(mat_contents['layers'][8])
conv35b = np.array(L1['biases'], dtype=np.float32)
conv35W = np.array(L1['filters'], dtype=np.float32).transpose((3,2,0,1))[:,:,::-1,::-1]
L1 = _todict(mat_contents['layers'][11])
conv4b = np.array(L1['biases'], dtype=np.float32)
conv4W = np.array(L1['filters'], dtype=np.float32).transpose((3,2,0,1))[:,:,::-1,::-1]
L1 = _todict(mat_contents['layers'][13])
dense1b = np.array(L1['biases'], dtype=np.float32)
dense1W = np.array(L1['filters'],dtype=np.float32)
dense1W = np.array(L1['filters'], dtype=np.float32).transpose((1,0,2,3))
dense1W = dense1W.reshape(4*13*512, 4096, order="F").copy()
L1 = _todict(mat_contents['layers'][15])
dense2b = np.array(L1['biases'], dtype=np.float32)
dense2W = np.array(L1['filters'],dtype=np.float32)
L1 = _todict(mat_contents['layers'][17])
classb = L1['biases']
classW = L1['filters']
np.savez_compressed('matlab_dictnet_weights',conv1W=conv1W,conv1b=conv1b,conv2W=conv2W,conv2b=conv2b,
conv3W=conv3W,conv3b=conv3b,conv35W=conv35W,conv35b=conv35b,conv4W=conv4W,
conv4b=conv4b,dense1W=dense1W,dense1b=dense1b,dense2W=dense2W,dense2b=dense2b,
classW=classW,classb=classb)
