The DALI 0.9.1 is
a beta release. Hence, for all the features, the functionality and performance will
likely be limited.
Using DALI 0.9.1 Beta
The DALI 0.9.1 can be used with the 19.05 NVIDIA GPU Cloud (NGC) optimized container for
MXNet, PyTorch, and TensorFlow. Also, the 19.05 container will be shipped with DALI
0.9.1.
To upgrade to DALI 0.9.1 beta from an older version of DALI, follow the installation
instructions in the DALI Quick Start Guide.
Refer to the DALI Developer Guide for usage details.
Note: The internal DALI C++ API used for operators implementation, and the C++ API that
enables using DALI as a library from native code, are not yet officially supported.
Hence these APIs may change in the next release without advance notice.
Key Features and Enhancements
This DALI release includes the following key features and
enhancements.
- Added Optical Flow example.
- Added full support for .pnm (.ppm/.pgm/.pbm) (in case of nvJPEG, fallback to CPU
is done).
- Added the ability to do the lazy initialization of DALI. It is now possible to
separate touching the data during the actual run, and at the time of all the
necessary allocations during the build of the pipeline.
- Reduced the resize operator scratch buffer size by processing the data in mini
batches.
- Added the ability to set the DALI CUDA streams priority.
- Fixed the sync issue in the DALI Python iterators.
- Fixed the initialization of CUDA context on the default device during pipeline
creation.
Breaking API Changes
- Internal Python pipeline API has changed. If any function
_* was used, then that function should be updated to reflect
the new semantic.