Deep Learning Inference for Fast Simulation Applications

C++ Inference Module for Generative TensorFlow Models

How To Run Inference

1. Follow the Inference Library instalation steps.

2. Run from an existing model (AR, GAN, VAE):

cd modules/
mkdir build
cd build
cmake3 .. -DCMAKE_PREFIX_PATH=<PATH_TO_INFERENCE_LIBRARY_INSTALL_PATH>
make 
./EventGeneration

How To Integrate Your Model

1. Save your input/output node names. For example, given a Python model:

# Event Data Innputs

x_sample = tf.placeholder(tf.float32, shape=xs, name='input_cells')
y_sample = tf.placeholder(tf.float32, shape=xs[0], name="input_labels")

# Generated Result

generation = tf.add(a, b, name='output_result')

2. Store the graph definition in a .pb file as well as the latest checkpoint in .ckpt files (.data, .index, .meta)

3. Note your input data shape information (both for samples and labels).

Model Integration Info File Example

modelType       =  "dcgan"
modelGraph      = "../dcgan.pb"
modelRestore    = "../model.b32.ckpt"
inputNode       = "input_cells"
labelNode       = "input_labels"
outputNode      = "output_result"
inputShape      = {64,8,8,24}
labelShape      = {64,100}