Problem with tensorflow serving

[dsabarinathan]

when I was creating the tensorflow serving model with frozen PB. I am getting an empty variable folder. could you share the un-frozen graph file?

Please help me on this issue.

[victordibia]

Which model are you referring to (v1 or v2)?
In the readme I have some directions on how to use the model checkpoints (which are available for mobilenetv1) to generate your own frozen graph.

In my experience using a frozen model from different TensorFlow models can be challenging. Its better to export using your own froze graph from checkpoints

[dsnsabari]

Hi Victordibia,

I am using ssdlitemobilenetv2 model checkpoint. Below I mentioned the code for converting the model checkpoint to serving. But empty variable folder was generated.

import tensorflow as tf

SAVE_PATH = D:/handtracking-master/model-checkpoint/ssdlitemobilenetv2/'
MODEL_NAME = 'test'
VERSION = 5
SERVE_PATH = './serve/{}/{}'.format(MODEL_NAME, VERSION)

checkpoint = tf.train.latest_checkpoint(SAVE_PATH)
print(checkpoint)

tf.reset_default_graph()

with tf.Session() as sess:
saver = tf.train.import_meta_graph(checkpoint + '.meta')
graph = tf.get_default_graph()
sess.run(tf.global_variables_initializer())
inputs = tf.saved_model.utils.build_tensor_info(graph.get_tensor_by_name('image_tensor:0'))
detection_boxes = tf.saved_model.utils.build_tensor_info(graph.get_tensor_by_name('detection_boxes:0'))
detection_scores = tf.saved_model.utils.build_tensor_info(graph.get_tensor_by_name('detection_scores:0'))
detection_classes = tf.saved_model.utils.build_tensor_info(graph.get_tensor_by_name('detection_classes:0'))
num_detections = tf.saved_model.utils.build_tensor_info(graph.get_tensor_by_name('num_detections:0'))

export_path =  './savedmodel/3'
builder = tf.saved_model.builder.SavedModelBuilder(export_path)


prediction_signature = (
  tf.saved_model.signature_def_utils.build_signature_def(
      inputs={'inputs': inputs},
      outputs={'output1': detection_boxes,'output2':detection_scores,'output3':detection_classes},
      method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME))

builder.add_meta_graph_and_variables(
  sess, [tf.saved_model.tag_constants.SERVING],
  signature_def_map={
      tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:
          prediction_signature 
  },
  )
builder.save()