Will there be a problem to load pretrained imagenet model?

[wanginwang]

It seems that in all your models such as dch.py, dhn.py, you first load model and pretrained model weights and then run global initializer,

self.img_last_layer, self.deep_param_img, self.train_layers, self.train_last_layer = self.load_model()

self.global_step = tf.Variable(0, trainable=False)
self.train_op = self.apply_loss_function(self.global_step)
self.sess.run(tf.global_variables_initializer())
return

We know that sess.run(tf.global_variables_initializer()) will initialize all trainable variables, so the model weights you load before are randomly initialized again if I understand correctly, right? Looking forward to your reply!!!