Home > Storage > PowerScale (Isilon) > Industry Solutions and Verticals > Analytics > PowerScale Deep Learning Infrastructure with NVIDIA DGX A100 Systems for Autonomous Driving > ADAS dataset and CNN models
Since the entire Cityscapes raw dataset, including label data, is only 11 GB and can easily fit into system memory. The size of the dataset was increased 557 times to exercise the storage system (Isilon F800) more realistically. We did this by applying random rotation data augmentation techniques to each JPEG image in the dataset. This is standard practice in data analytics and DL to increase size of data sets. In total, this “557x” dataset contained 2,025,975 images and annotations with 5.4 TB in total. The average PNG image pixel is 1024 x 2048 and average size is 2,193KB.
The performance test utilized in this document uses this data to train two different convolutional neural network (CNN) models, shown in the following table, that are used for semantic image segmentation and object detection.
CNN model |
Purpose |
Dataset |
DL framework |
Distributed framework |
Real-Time Object Detection |
Cityscapes Benchmark data with 972,825 training images (2.8 TB in total) |
PyTorch 1.4 |
PyTorch Distributed Data Parallel |
|
Hierarchical Multi-Scale Attention for Semantic Segmentation. |
Cityscapes Benchmark data with 2,025,975 training images (5.4 TB in total) |
PyTorch 1.3 |
PyTorch Distributed Data Parallel |