Reputation: 1
I was following this link to train audio recognition model. https://www.tensorflow.org/versions/master/tutorials/audio_recognition
Started with the command:
python tensorflow/examples/speech_commands/train.py
End of the command shows :
INFO:tensorflow:Saving to "/media/slattann/HDD2/tmp/speech_commands_train/conv.ckpt-18000"
INFO:tensorflow:Confusion Matrix:
[[257 0 0 0 0 0 0 0 0 0 0 0] [ 0 194 4 3 4 7 8 16 5 0 7 9] [ 0 4 239 1 3 0 6 2 0 0 1 0] [ 1 4 0 216 1 6 3 4 1 0 3 13] [ 0 0 0 0 257 0 2 1 3 1 6 2] [ 2 7 0 13 3 213 2 0 1 0 1 11] [ 0 2 11 0 5 0 246 3 0 0 0 0] [ 1 11 0 1 1 1 0 240 1 2 1 0] [ 1 7 0 0 1 0 1 2 234 0 0 0] [ 1 4 0 0 19 0 1 2 9 222 3 1] [ 0 2 1 0 7 1 1 0 1 0 235 1] [ 0 6 0 33 5 3 6 2 0 0 2 194]]
INFO:tensorflow:Final test accuracy = 89.2% (N=3081)
but i could not find this checkpoint file. at the end I could only see:
----------------------------------------------------------------------------
Files I could see in the /media/slattann/HDD2/tmp/speech_commands_train folder:
checkpoint
conv.ckpt-17600.data-00000-of-00001
conv.ckpt-17600.index
conv.ckpt-17600.meta
conv.ckpt-17700.data-00000-of-00001
conv.ckpt-17700.index
conv.ckpt-17700.meta
conv.ckpt-17800.data-00000-of-00001
conv.ckpt-17800.index
conv.ckpt-17800.meta
conv.ckpt-17900.data-00000-of-00001
conv.ckpt-17900.index
conv.ckpt-17900.meta
conv.ckpt-18000.data-00000-of-00001
conv.ckpt-18000.index
conv.ckpt-18000.meta
conv_labels.txt
conv.pbtxt
----------------------------------------------------------------------------
I am blocked to proceed with freezing the graph stage before creating the tensorflow model(.pb).
1: https://i.sstatic.net/TIXZX.jpg
Upvotes: 0
Views: 557
Reputation: 19123
When it says Saving to "/media/slattann/HDD2/tmp/speech_commands_train/conv.ckpt-18000"
what it really means is that it creates (at least) three files with the same prefix:
To freeze_graph
just pass the prefix path/to/checkpoint/dir/conv.ckpt-18000
, it will handle the rest on its own.
For more informatoin about why there are three files in the first place, see this question
Upvotes: 1