DALI Release 0.4 Beta
Key Features and Enhancements
This DALI release includes the following key features and
enhancements.
- Added basic operators for detection (SSD), such as COCO dataset reader (COCOReader), random crop operator with bounding boxes (SSDRandomCrop), and flip operator for bounding boxes (BbFlip).
- Added CPU versions of Crop/CropCastPermute operators.
- Added a random Paste operator.
- Upgraded OpenCV to 3.4.3.
- Upgraded MXNet to 1.3.0.
- Fixed parsing of JPEG headers by the Host Decoder known issue. The Host Decoder now handles all images from the ImageNet dataset.
- Added fine grained control over output buffers in the pipeline
- Updated to nvJPEG 0.2.0
Breaking API Changes
The pipeline constructor signature has changed; an additional argument (prefetch_queue_depth) was added to allow defining depth of the prefetch queue at runtime.
Using DALI 0.4 Beta
The 18.09 NVIDIA GPU Cloud (NGC) optimized container for MXNet, PyTorch, and TensorFlow, includes an older version of DALI. To upgrade to DALI 0.4 beta, follow the installation instructions in the DALI Quick Start Guide.
Refer to the DALI Developer Guide for usage details.
Deprecated Features
- DALI 0.4 is not compatible with TensorFlow 1.11. This will be addressed in the next release.
Known Issues
- The DALI integrated ResNet-50 samples in the 18.10 NGC TensorFlow and PyTorch containers may result in lower than expected performance results. We are working to address the issue in the next release.
- This is a beta release. All features are expected to be available, however, some aspects of functionality and performance will likely be limited compared to a non-beta release.