1

I used the Inception-library (Tensorflow/models/Inception-tutorial on github), feeding selfmade TFRecord-files to imagenet_train.py, and ended up with a 7Gb directory containing .ckpt and .meta files.

How do i use them to make actual predictions?(also evaluating, testing, testing with actual .jpg files, continue the training)

What would be the next steps?Are there dummy-friendly examples(tutorials)?

[I know there´s a thread with a similar name, but it contains barely a question nor answers.]

Michael Bauer
  • 21
  • 4

1 Answers1

0

https://www.tensorflow.org/versions/r0.10/how_tos/variables/index.html#restoring-variables provides pretty much everything you should need.

The code there is:

# Add ops to save and restore all the variables.
saver = tf.train.Saver()

# Later, launch the model, use the saver to restore variables from disk, and
# do some work with the model.
with tf.Session() as sess:
  # Restore variables from disk.
  saver.restore(sess, "/tmp/model.ckpt")
  print("Model restored.")
  # Do some work with the model
  ...

So at the ... part, you could load your .jpegs and feed them into the network with placeholders and feeddicts.

Jacob Holloway
  • 887
  • 8
  • 24
  • Let me know if you need more help. This is more of a bare-bone's answer to get you started. – Jacob Holloway Sep 05 '16 at 15:35
  • 1
    The "..." is actually that gives me the most headaches. Is there newbie-friendly example code, that works without prebuild data-sets? – Michael Bauer Sep 05 '16 at 17:59
  • Push. Help please. I keep on struggling with using my Checkpoint files, but really need them to finish my school-project. – Michael Bauer Sep 11 '16 at 16:30
  • Oh, sorry Michael. Life got busy. Graduation, new job, baby, house hunting...I'll set some time aside tomorrow to help you out, probably around lunch. – Jacob Holloway Sep 12 '16 at 00:23
  • Thank you very much. Congratulation, and all the best. – Michael Bauer Sep 13 '16 at 20:11
  • Anything new yet? – Michael Bauer Sep 20 '16 at 13:29
  • I'm really sorry Michael, Life is so busy. :( I don't even have time to do my own experiments and that makes me sad. It makes me feel worse that I don't have time to help out like I said. The best I can do is send you to this one maybe: http://stackoverflow.com/questions/33759623/tensorflow-how-to-restore-a-previously-saved-model-python/ Maybe copy down their files down, and add in a train flag, then add in a if train (do their stuff) else eval (do your stuff.) It shouldn't be terrible to figure out (I hope. :) ) – Jacob Holloway Sep 20 '16 at 15:40
  • Again, sorry about offering to help and then not being able to. I do feel quite bad about that. – Jacob Holloway Sep 20 '16 at 15:40