DIGITS Release 5.0

Key Features and Enhancements

The DIGITS version 5.0 release includes the following key features and enhancements. These enhancements are new since DIGITS version 4.0.

• Enabled the DIGITS Model Store
• Fixed calculations related to batch accumulation for NVCaffe™
• Support for image segmentation workflows
• Online data augmentation with Torch™
• Show CPU and system memory utilization during training
• Improved bounding-box visualizations for object detection models
• Created groups of jobs for easier display on the home page
• Enabled sharing of data visualization extensions for both dataset exploration and viewing inference results
• Support for plugin extensions
• Added documentation for the REST API
• Fixed bug with Torch and CUDA_VISIBLE_DEVICES
• Fixed issues with browsers returning incorrectly cached css and js files
• New design for Torch multi-GPU training
• Added Ubuntu 16.04 support by updating dependency versions
• digits-devserver is now a small shell script instead of a Python script

Using DIGITS 5.0

Ensure you are familiar with the following notes when using this version.

• Use environment variables for configuration instead of a file

Deprecated Features

• Removed the digits-server script and its dependency on gunicorn

Known Issues

• Training goes on longer than required when using batch accumulation