Facenet: src/classifier.py Not Taking ckpt_file file or meta_file AND src/train_softmanx.py DOES NOT generate .pb (protobuf) file

[akshaydotsh]

After training Inception Resnet V1 on my custom images using train_softmax.py produces such directory structure :

20180827-104410  20180827-115022  20180827-120731  20180827-121107  20180827-121544  20180827-122058  20180827-122452
20180827-114425  20180827-120607  20180827-120952  20180827-121259  20180827-121920  20180827-122206  20180827-123746

In this arrangement only the last folder 20180827-123746 has checkpoint and meta files like so :

checkpoint                                         model-20180827-123746.ckpt-49.data-00000-of-00001  model-20180827-123746.ckpt-50.index
model-20180827-123746.ckpt-48.data-00000-of-00001  model-20180827-123746.ckpt-49.index                model-20180827-123746.meta
model-20180827-123746.ckpt-48.index                model-20180827-123746.ckpt-50.data-00000-of-00001

So the problem is I am not getting any .pb file to feed to src/classifier.py
Although it is mentioned in classifier.py file that it takes ckpt_file and meta_file but it is giving the following error

Traceback (most recent call last):
  File "src/classifier.py", line 170, in <module>
    main(parse_arguments(sys.argv[1:]))
  File "src/classifier.py", line 69, in main
    facenet.load_model(args.model)
  File "/home/ubuntu/facenet/src/facenet.py", line 372, in load_model
    graph_def.ParseFromString(f.read())
google.protobuf.message.DecodeError: Error parsing message

I don't know any other approach to make this work. Can you help with any alternate solution ?
I tried running classifier directly on 20180408-102900.pb BUT surprising it gave me accuracy of 0.129

[rgsousa88]

To generate the .pb file, you have to run freeze_graph.py script. This script have two positional arguments: the directory with metagraph and checkpoint files and the output file path where you want the model to be saved.