Home > Storage > PowerScale (Isilon) > Industry Solutions and Verticals > Analytics > Deep Learning with Dell EMC Isilon > Introduction
In this section, the performance of deep learning training is measured using TensorFlow. The well-known ILSVRC2012 dataset (often referred to as ImageNet) was used for benchmarking performance. This dataset contains 1,281,167 training images in 144.8 GB. (All unit prefixes use the SI standard—base 10—where 1 GB is 1 billion bytes.) All images are grouped into 1000 categories or classes. This dataset is commonly used by deep learning researchers for benchmarking and comparison studies.
Since the entire ILSVRC2012 dataset is only 144.8 GB and can easily fit into system memory, the size of the dataset was increased 10 times to properly exercise the storage system (Isilon F800). We did this by applying 10 random data augmentation techniques to each JPEG image in the dataset. This is standard practice in data analytics and deep learning to increase size of data sets. The augmented images were then converted to a TensorFlow TFRecords database. In total, this “10x” dataset contained 12,811,670 JPEG images in 1,448 GB split across 1024 files. The average JPEG image size is 113 KB, and the average TFRecord file size is 1.41 GB.
TFRecords file format is a Protocol Buffers binary format that combines multiple raw image files together with their metadata into one binary file. It maintains the image compression offered by the JPEG format and the total size of the dataset remained the same.
The benchmark utilized in this document uses this data to train two different convolutional neural network (CNN) models that are popular for image classification – ResNet50 and AlexNet. ResNet50 is much more accurate but it requires a lot more computation. Although neither of these are state-of-the-art today, knowing their performance can be useful when evaluating different hardware and software platforms.