White Papers

12 CheXNet – Inference with Nvidia T4 on Dell EMC PowerEdge R7425 | Document ID
'Effussion',
'Hernia',
'Nodule',
'Pneumonia',
'Atelectasis',
'PT',
'Mass',
'Edema',
'Consolidation',
'Infiltration',
'Fibrosis',
'Pneumothorax']
Build a Convolutional Neural Network using Estimators:
Here we describe the building process of the CheXNet model with Transfer Learning using
Custom Estimator. We used the high-level TensorFlow API tf.estimator and its class
Estimator to build the model, it handles the high-level model training, evaluation, and
inference of our model much easier than with the low-level TensorFlow APIs; it builds the
graph for us and simplifies sharing the implementation of the model on a distributed multi-
server environment, among other advantages.[3].
There are pre-made estimators and custom estimators [4], in our case we used the last one
since it allows to customize our model through the model_fn function. Also, we defined the
input_fn function which provides batches for training, evaluation, and prediction. When the
tf. estimator class is called, it returns an initialized estimator, that at the same time calls the.
train, eval, and predict functions, handling graphs and sessions for us.
See Figure 6 with the overview of the estimator.