how to continue to train model

[aa642531]

I have trained my model.I want to continue to train it.How to do change the code?

[cemysf]

In the train_model function of SolverWrapper class in lib/fast_rcnn/train.py: Initialize variables with "self.saver.restore(sess, \<path of ckpt file>), instead of "self.net.load" Also, you may want to set the "iter" count down below to its new correct value

You can add the path of ckpt file in SolverWrapper class as a parameter and pass it from tools/train_net.py

[xinboli]

@aa642531 I have the same question with you, do you solve the problem? @cemysf I tried the method as you said, like this in the train.py:

iintialize variables

    sess.run(tf.global_variables_initializer())
    if self.pretrained_model is not None:
        print ('Loading pretrained model '
               'weights from {:s}').format(self.pretrained_model)
        #self.net.load(self.pretrained_model, sess, self.saver, True)
        self.saver.restore(sess, VGGnet_fast_rcnn_iter_10.ckpt) #this ismy new ckpt file

but I got this error: File "/home/xinboli/work_code/tmp/Faster-RCNN_TF/tools/../lib/fast_rcnn/train.py", line 157, in train_model self.saver.restore(sess, VGGnet_fast_rcnn_iter_10.ckpt) NameError: global name 'VGGnet_fast_rcnn_iter_10' is not defined

Would you like to help me? thank you very much!

[cemysf]

@xinboli you should give the path as a string, like this: self.saver.restore(sess, 'VGGnet_fast_rcnn_iter_10.ckpt')

[xinboli]

@cemysf ahhhhhh! I know!! Now i can use it, thank you :)