operations.py 文件源码

def load_weights(model, sess, weight_file):
  """
  Load weights from given weight file (used to load pretrain weight of vgg model)

  Args:
    model            :         model to restore variable to
    sess             :         tensorflow session
    weight_file      :         weight file name
  """

  weights = np.load(weight_file)
  keys    = sorted(weights.keys())
  for i, k in enumerate(keys):
    if i <= 29:
      print('-- %s %s --' % (i,k))
      print(np.shape(weights[k]))
      sess.run(model.parameters_conv[i].assign(weights[k]))