Planning for PowerAI Vision

You must meet the software and hardware requirements and understand the supported file types before you can install PowerAI Vision.

Hardware requirements

PowerAI Vision requires the following hardware:
  • POWER8 S822LC (8335-GTB) or POWER9 AC922 with at least one NVIDIA NVLink capable GPU
  • 128 GB of memory
  • Ethernet network interface
  • 40 GB of storage. See Disk space requirements for details.
  • If Optimized for speed (tiny YOLO v2) is selected when training the model, there are multiple options for deploying the model for testing. Deploying a model to a Xilinx FPGA requires the Xilinx Alveo U200 Accelerator card.

Software requirements

You must install the following software before you install PowerAI Vision:
Linux
  • Red Hat Enterprise Linux (RHEL) 7.6 (little endian).
  • Ubuntu 18.04 or later.
NVIDIA CUDA
10.1 Update 1 or later drivers. For information, see the NVIDIA CUDA Toolkit website.
Docker
  • Docker must be installed. The recommended version is 1.13.1 or later. Version 1.13.1 is installed with RHEL 7.6.
  • Ubuntu - Docker CE or EE 18.06.01

Networking requirements

Your environment must meet the following networking requirements:
  • A default route must be specified on the host system.
    • For instructions to do this on Ubuntu, refer to the IP addressing section in the Ubuntu Network Configuration. Search for the steps to configure and verify the default gateway.
    • For instructions to do this on Red Hat Enterprise Linux (RHEL), refer to 2.2.4 Static Routes and the Default Gateway in the Red Hat Customer Portal.
  • For RHEL, Docker 0 must be in a trusted firewall zone. If it is not in a trusted firewall zone, modify the RHEL settings as follows: 
    sudo nmcli connection modify docker0 connection.zone trusted
sudo systemctl stop NetworkManager.service
sudo firewall-cmd --permanent --zone=trusted --change-interface=docker0
sudo systemctl start NetworkManager.service
sudo nmcli connection modify docker0 connection.zone trusted
sudo systemctl restart docker.service
  • IPv4 port forwarding must be enabled.

    If IPv4 port forwarding is not enabled, run the /sbin/sysctl -w net.ipv4.conf.all.forwarding=1 command. For more information about port forwarding with Docker, see UCP requires IPv4 IP Forwarding in the Docker success center.

  • IPv6 must be enabled.

Disk space requirements

PowerAI Vision has the following storage requirements for the initial product installation and for the data sets that will be managed by the product.

Standalone installation
  • /var - The product installation requires at least 25 Gb of space in the /var file system for the product Docker images. PowerAI Vision also generates log information in this file system.

    Recommendation: If you want to minimize the root (/) file system, make sure that /var has its own volume. The /var file system should have at least 50 Gb of space, more if additional applications are being run on the system that use this file system for log data and so on.

  • /opt - PowerAI Vision data sets are stored in this file system. The storage needs will vary depending on the data sets and the contents - i.e., video data can require large amounts of storage.

    Recommendation: If you want to minimize the root (/) file system, make sure that /opt has its own volume. The /opt file system should have at least 40 Gb of space, although this value might be more depending on your data sets.

IBM Cloud Private installation
The PowerAI Vision product will use the configured persistent storage for the deployment, the requirements are documented in Installing PowerAI Vision with IBM Cloud Private.

Supported web browsers

The following web browsers are supported:

  • Google Chrome Version 60, or later
  • Firefox Quantum 59.0, or later

Image support

  • The following image formats are supported:
    • JPEG
    • PNG
  • Images with COCO annotations are supported. For details, see Importing images with COCO annotations.
  • PowerAI Vision has limited support for Pascal VOC annotations. Annotations for multiple files residing in a common XML file are not supported. In other words, each annotation XML file can only contain annotations for a single image, identified by the filename attribute.

    If you have a single XML annotation file containing annotations for multiple images in the data set to be imported, the annotations need to be split out into separate XML files before PowerAI Vision can import the annotations successfully.

  • The models used by PowerAI Vision have limitations on the size and resolution of images. If the original data is high resolution, then the user must consider:
    • If the images do not need fine detail for classification or object detection, they should be down-sampled to 1-2 megapixels.
    • If the images do require fine detail, they should to be divided into smaller images of 1-2 megapixels each.
    • High resolution images will be scaled to a maximum of 1000 x 600 pixels.
    • For image classification, images are scaled to 224 x 224 pixels.
    • For object detection with Detectron, all images are scaled to 1333 x 800 pixels. However, the original aspect ratio is maintained. If necessary, black bands are added to the image to make it fit.
    • For object detection with tiny YOLO V2, all images are scaled to 416 x 416. The longest edge is scaled to 416 pixels and, if necessary, black bands are added to the shorter side to make it 416 pixels.
    • For object detection with FR-CNN and video action detection, anything over 1000 x 600 pixels is down-sampled so that the longest edge will fit. However, the original aspect ratio is maintained. If necessary, black bands are added to the image to make it fit.
    • There is a 24 GB size limit per upload session. This limit applies to a single .zip file or a set of files. You can, however upload 24 GB of files, then upload more after the original upload completes.

Supported video types

The following video formats are supported:
Can be played in the PowerAI Vision GUI:
  • Ogg Vorbis (.ogg)
  • VP8 or VP9 (.webm)
  • H.264 encoded videos with MP4 format (.mp4)
Supported by API only:
  • Matroska (.mkv)
  • Audio Video Interleave (.avi)
  • Moving Picture Experts Group (.mpg or .mpeg2)
Not supported:
Videos that are encoded with the H.265 codec.

Deep learning frameworks

The following frameworks are included with PowerAI Vision.

Table 1. Included frameworks
Framework Version Python 2.7 support Python3.6 support Notes
Caffe 2 1.0.0 Yes No Supported for Detectron models
IBM® Caffe 1.0 Yes No Supported for GoogLeNet, Faster R-CNN, and tiny YOLO V2 models
Keras 2.2.4 Yes Yes Supported for custom models
TensorFlow 1.14 Yes Yes Supported for custom models

Limitations

Following are some limitations for IBM PowerAI Vision 1.1.4:
  • PowerAI Vision uses an entire GPU when you are training a dataset. Multiple GoogleNet or Faster R-CNN models can be deployed to a single GPU. Other types of models take an entire GPU when deployed. For details about other differences between model types, see Model functionality.

    The number of active GPU tasks (model training and deployment) that you can run at the same time depends on the number of GPUs on your Power® System server. You must verify that there are enough available GPUs on the system for the desired workload. The number of available GPUs is displayed on the user interface.

  • You cannot install PowerAI Vision stand-alone on a system that already has any of these products installed:
    • IBM Data Science Experience (DSX)
    • IBM Cloud Private
    • Any other Kubernetes based applications