I have a Keras model I converting to a tensorflow serving model. I can successfully convert my pretrained keras model to take b64 input, preprocess that input, and feed it to my model. My problem is that I don't know how to take the prediction data I am getting (which is enormous) and only export the top result. I am doing image segmentation so my output prediction is of shape (?, 473, 473, 3) and I'd like to get the top result and return it in b64 encoded format. What I have currently that just returns the entire prediction:
sess = K.get_session()
g = sess.graph
g_def = graph_util.convert_variables_to_constants(sess,
g.as_graph_def(),
[model.output.name.replace(':0','')])
with tf.Graph().as_default() as g_input:
input_b64 = tf.placeholder(shape=(1,),
dtype=tf.string,
name='b64')
tf.logging.info('input b64 {}'.format(input_b64))
image = tf.image.decode_image(input_b64[0])#input_bytes)
image_f = tf.image.convert_image_dtype(image, dtype=tf.float16)
input_image = tf.expand_dims(image_f, 0)
image_r = tf.image.resize_bilinear(input_image, [HEIGHT, WIDTH], align_corners=False)
input_data = preprocess_image(image_r)
output = tf.identity(input_data, name='input_image')
# Convert to GraphDef
g_input_def = g_input.as_graph_def()
with tf.Graph().as_default() as g_combined:
x = tf.placeholder(tf.string, name="b64")
im, = tf.import_graph_def(g_input_def,
input_map={'b64:0': x},
return_elements=["input_image:0"])
pred, = tf.import_graph_def(g_def,
input_map={model.input.name: im},
return_elements=[model.output.name])
with tf.Session() as session:
inputs = {"image_bytes": tf.saved_model.utils.build_tensor_info(x)}
outputs = {"output_bytes":tf.saved_model.utils.build_tensor_info(pred)}
signature =tf.saved_model.signature_def_utils.build_signature_def(
inputs=inputs,
outputs=outputs,
method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME
)
"""Convert the Keras HDF5 model into TensorFlow SavedModel."""
if os.path.exists(export_path):
shutil.rmtree(export_path)
legacy_init_op = tf.group(tf.tables_initializer(), name='legacy_init_op')
builder = saved_model_builder.SavedModelBuilder(export_path)
builder.add_meta_graph_and_variables(
sess=session,
tags=[tag_constants.SERVING],
signature_def_map={ signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY: signature },
)
builder.save()
I took a lot of what I have working from https://medium.com/google-cloud/serverless-transfer-learning-with-cloud-ml-engine-and-keras-335435f31e15 for reference. Thanks!
Posting my own solution I made in case someone else runs into this issue. Basically, you just do the inverse of the input function.