tf_vs_trt_benchmark.py

import tensorflow as tf
import tensorflow.contrib.tensorrt as trt
import numpy as np
import PIL
from timeit import default_timer as timer
from tqdm import tqdm


'''
This script benchmarks how long it takes to run perform inference on a pure Tensorflow (TF) model vs a converted TensorRT model
In order to run it, it assumes you have a frozen TF graph in the form of a *.pb file.

The specific model is MobileNetv2 
(https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet(
Frozen graph can be downloaded from: https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.4_224.tgz
'''

## File paths (needs to be set)
pb_path = "mobilenet_v2_1.0_224_frozen.pb"    # frozen TF graph
input_name = 'input'                                    # input layer of the graph
output_names = ['MobilenetV2/Predictions/Reshape_1']    # output layer of the graph. (Can be multiple)
img_path = 'data/panda.jpg'     # image to perform inference on

## Timing paramters
trls = 100   # number of inference trials to average over

## Load image
img_raw = np.array(PIL.Image.open(img_path).resize((224, 224))).astype(np.float) / 128 - 1
img = img_raw.reshape(1, 224,224, 3)

##### Run w/ standard TF
frozen_graph = tf.GraphDef.FromString(open(pb_path, 'rb').read())
inp, predictions = tf.import_graph_def(frozen_graph,  return_elements = ['input:0', 'MobilenetV2/Predictions/Reshape_1:0'])
with tf.Session(graph=inp.graph):
    start = timer()
    tf_trl_time = []
    for _ in tqdm(range(trls)):
        x = predictions.eval(feed_dict={inp: img + np.random.random(img.shape)/10})
        tf_trl_time += [timer() - start]
tf_avg_trl_time = np.mean(np.diff(tf_trl_time))
print("TF: Top 1 prediction: ", x.argmax(), x.max())
print('TF took on average %.3fs over %d trls' % (tf_avg_trl_time, trls))

##### Run w/ TensorRT
trt_graph = trt.create_inference_graph(
        input_graph_def=frozen_graph,
        outputs=output_names,
        max_batch_size=1,
        precision_mode='FP16',  # 'INT8'/'FP16'
)

tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
tf_sess = tf.Session(config=tf_config)
tf.import_graph_def(trt_graph, name='')
tf_input = tf_sess.graph.get_tensor_by_name(input_name + ':0')
tf_output = tf_sess.graph.get_tensor_by_name(output_names[0] + ':0')

start = timer()
trt_trl_time = []
for _ in tqdm(range(trls)):
    output = tf_sess.run(tf_output, feed_dict={tf_input: img + np.random.random(img.shape)/10})
    trt_trl_time += [timer() - start]
trt_avg_trl_time = np.mean(np.diff(trt_trl_time))
print("TRT: Top 1 prediction: ", output.argmax(), output.max())
print('TRT took on average %.3fs over %d trls' % (trt_avg_trl_time, trls))