Error loading model using downloaded files

[DGM1193]

Thanks for sharing your work ! Looks very neat .

I am trying to load the model provided and test it on my images. I get this error when running the eval.py script.
"InvalidArgumentError (see above for traceback): Restoring from checkpoint failed. This is most likely due to a mismatch between the current graph and the graph from the checkpoint. Please ensure that you have not altered the graph expected based on the checkpoint. Original error:

Cannot assign a device for operation model_0/split: ...."

Could you please let me know what the issue is?

[liuheng92]

the graph from the checkpoint is as same as the graph from the code. I am sure about it. Maybe you can check your setting and running environment

[DGM1193]

Thanks for the quick reply ! I solved the previous error I had, and I get this error now :

line 171, in main
model_path =os.path.join(FLAGS.checkpoint_path,os.path.basename(ckpt_state.model_checkpoint_path))
AttributeError: 'NoneType' object has no attribute 'model_checkpoint_path'

[liuheng92]

create a checkpoint file yourself

[DGM1193]

I solved this issue by replacing lines 169 to 173 in eval.py which is :

with tf.Session(config=tf.ConfigProto(allow_soft_placement=True)) as sess:
        ckpt_state = tf.train.get_checkpoint_state(FLAGS.checkpoint_path)
        model_path = os.path.join(FLAGS.checkpoint_path,os.path.basename(ckpt_state.model_checkpoint_path))
        logger.info('Restore from {}'.format(model_path))
        saver.restore(sess, model_path)

with the following lines of code:

with tf.Session(config=tf.ConfigProto(allow_soft_placement=True)) as sess:
    saver = tf.train.Saver()
    saver.restore(sess, model_path)    # path to checkpoint file - I gave this as "~/checkpoint/model.ckpt "even though I did not have a checkpoint file. the directory "~/checkpoint/" consists of the 3 files which can be downloaded from this repo, thanks to @liuheng92 .

and using tensorflow version 1.8