Exporting model using SavedModelBuilder about rude-carnie

[dpressel]

I will take a look when I get a second. Sorry, am quite busy ATM.

[dpressel]

I added an exporter for TF serving as you suggested:

3868a1a

This should also resolve the questions you had about how to export it to TF serving.

To run it with tensorflow serving steps I did:

First export the model:

dpressel@dpressel:~/dev/work/rude-carnie$ python export.py --model_type inception --model_dir /data/xdata/rude-carnie/checkpoints/age/inception/22801 --output /tmp/age
2017-06-21 11:45:30.503473: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:893] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2017-06-21 11:45:30.503847: I tensorflow/core/common_runtime/gpu/gpu_device.cc:906] Found device 0 with properties: 
name: GeForce GTX 980M
major: 5 minor: 2 memoryClockRate (GHz) 1.1265
pciBusID 0000:01:00.0
Total memory: 7.93GiB
Free memory: 7.09GiB
2017-06-21 11:45:30.503883: I tensorflow/core/common_runtime/gpu/gpu_device.cc:927] DMA: 0 
2017-06-21 11:45:30.503889: I tensorflow/core/common_runtime/gpu/gpu_device.cc:937] 0:   Y 
2017-06-21 11:45:30.503895: I tensorflow/core/common_runtime/gpu/gpu_device.cc:996] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce GTX 980M, pci bus id: 0000:01:00.0)
selected (fine-tuning) inception model
/data/xdata/rude-carnie/checkpoints/age/inception/22801/checkpoint-14999
Restored model checkpoint /data/xdata/rude-carnie/checkpoints/age/inception/22801/checkpoint-14999
Exporting trained model to b'/tmp/age/1'
WARNING:tensorflow:Error encountered when serializing LAYER_NAME_UIDS.
Type is unsupported, or the types of the items don't match field type in CollectionDef.
'dict' object has no attribute 'name'
Successfully exported model to /tmp/age

The exporter is based on:

https://github.com/tensorflow/serving/blob/master/tensorflow_serving/example/inception_client.py

Now serve the model from tensorflow_serving directory:

dpressel@dpressel:~/dev/work/serving$ bazel-bin/tensorflow_serving/model_servers/tensorflow_model_server --port=9000 --model_name=age --model_base_path=/tmp/age/
...
2017-06-21 11:45:58.713786: I external/org_tensorflow/tensorflow/cc/saved_model/loader.cc:146] Restoring SavedModel bundle.
2017-06-21 11:45:59.194215: I external/org_tensorflow/tensorflow/cc/saved_model/loader.cc:181] Running LegacyInitOp on SavedModel bundle.
2017-06-21 11:45:59.249601: I external/org_tensorflow/tensorflow/cc/saved_model/loader.cc:275] Loading SavedModel: success. Took 814539 microseconds.
2017-06-21 11:45:59.249658: I tensorflow_serving/core/loader_harness.cc:86] Successfully loaded servable version {name: age version: 1}
2017-06-21 11:45:59.263568: I tensorflow_serving/model_servers/main.cc:298] Running ModelServer at 0.0.0.0:9000 ...

Here is a sample client to run:

https://gist.github.com/dpressel/6d77c89203576ac3d5e35237c0f4f94f

dpressel@dpressel:~/dev/work/tf-serving-py$ python client.py --image ~/Downloads/portraits/biebs.jpg 
outputs {
  key: "classes"
  value {
    dtype: DT_STRING
    tensor_shape {
      dim {
        size: 1
      }
      dim {
        size: 2
      }
    }
    string_val: "(15, 20)"
    string_val: "(25, 32)"
  }
}
outputs {
  key: "scores"
  value {
    dtype: DT_FLOAT
    tensor_shape {
      dim {
        size: 1
      }
      dim {
        size: 2
      }
    }
    float_val: 0.8679360747337341
    float_val: 0.06530137360095978
  }
}

Note that running this is equivalent to running guess.py with --single_look (it doesnt do the multi-crop approach which is the default in guess.py)