NVIDIA/DALI v1.36.0 on GitHub

Key Features and Enhancements

This DALI release includes the following key features and enhancements:

Added support for checkpointing in MXNet iterator and CPU TensorFlow plugin (#5334, #5315).
Added morphological operators (fn.experimental.dilate, fn.experimental.erode) (#5294).
Integrated nvImageCodec for decoding in fn.experimental.decoders (#5297, #5336, #5324, #5333, #5339).
Added fn.random_crop_generator operator (#5304).
Added support for multiple inputs and relative shapes and anchors in fn.multi_paste (#5331).

Fixed Issues

Fixed insufficient synchronization in MXNet iterator (#5364).
Fixed auto_reset argument handling in iterator plugins (#5340).
Fixed missing calls to nvml::Shutdown (#5317).
Limited a number of progressive scans for jpeg decoding (#5316).

Improvements

Propagate module and display name of the operator to backend (#5344)
Update dependencies (#5349)
Map backend exceptions into Python exception types (#5345)
Emphasise the optical flow is calculated at input resolution. (#5350)
Refactor custom ops classes to use python_op_factory as base (#5338)
Add origin stack trace capture for DALI operators (#5302)
Test fused decoder with two separate pipelines (#5343)
[Cutmix] Make fn.multi_paste more flexible, fix validation (#5331)
Enable checkpointing in TensorFlow plugin (CPU only) (#5334)
Copy out nvImageCodec conda package from the build (#5336)
Add error message when GPU is not available (#5329)
Enable build with statically linked nvimgcodec + hard dependency for dynamic linking (#5324)
Add tf_stack util to autograph (#5322)
Rewrite median blur to use nvcvop tools (#5327)
Add morphological operators and the nvcvop module (#5294)
Add OpSpec::ArgumentInputIdx (#5330)
Simplify workspace object. Ensure predictable argument order in OpSpec. (#5325)
Support checkpointing in MXNet iterator (#5315)
Set rpath at cmake level (do not wait for bundle-wheel) (#5323)
Interpolation modes documentation upgrade (#5314)
Update links in DALI documentation (#5321)
Integrate nvimagecodec (#5297)
Add naive_histogram custom operator to test suite (#4731)
Add RandomCropGenerator (#5304)
Use small videos in checkpointing tests (#5305)

Bug Fixes

Use synchronous copy to framework array in the absence of a stream (#5364)
Process TFRecord reader binding classes only when it is enabled (#5360)
Adjust stack formatting in backend to match Python (#5354)
Link test operators against nvml wrapper (#5355)
Fix range check in Workspace::SetInput (#5358)
Make async_pool immune to stream handle reuse. (#5348)
Coverity fixes for 1.36 (#5342)
Fix "auto_reset" argument handling (#5340)
Fix cupy tests (#5341)
Add nvimagecodec libs to DALI_EXCLUDES + test utils to dump mismatched images (#5339)
Fix warning about nvImageCodec version (#5333)
Silence warning about DOWNLOAD_EXTRACT_TIMESTAMP while fixing the cmake <3.24 builds (#5326)
Fix inconsistent calls to nvml::Init and nvml::Shutdown (#5317)
Limit the number of progressive scans for jpeg decoding (#5316)

Breaking API changes

There are no breaking changes in this DALI release.

Deprecated features

No features were deprecated in this release.

Known issues:

The following operators: experimental.readers.fits, experimental.decoders.video, experimental.inputs.video, and experimental.decoders.image_random_crop do not currently support checkpointing.
The video loader operator requires that the key frames occur, at a minimum, every 10 to 15 frames of the video stream.
If the key frames occur at a frequency that is less than 10-15 frames, the returned frames might be out of sync.
Experimental VideoReaderDecoder does not support open GOP.
It will not report an error and might produce invalid frames. VideoReader uses a heuristic approach to detect open GOP and should work in most common cases.
The DALI TensorFlow plugin might not be compatible with TensorFlow versions 1.15.0 and later.
To use DALI with the TensorFlow version that does not have a prebuilt plugin binary shipped with DALI, make sure that the compiler that is used to build TensorFlow exists on the system during the plugin installation. (Depending on the particular version, you can use GCC 4.8.4, GCC 4.8.5, or GCC 5.4.)
In experimental debug and eager modes, the GPU external source is not properly synchronized with DALI internal streams.
As a workaround, you can manually synchronize the device before returning the data from the callback.
Due to some known issues with meltdown/spectra mitigations and DALI, DALI shows best performance when running in Docker with escalated privileges, for example:
- privileged=yes in Extra Settings for AWS data points
- --privileged or --security-opt seccomp=unconfined for bare Docker.

Binary builds

NOTE: DALI builds for CUDA 12 dynamically link the CUDA toolkit. To use DALI, install the latest CUDA toolkit.

CUDA 11.0 and CUDA 12.0 builds use CUDA toolkit enhanced compatibility. 
They are built with the latest CUDA 11.x/12.x toolkit respectively but they can run on the latest, 
stable CUDA 11.0/CUDA 12.0 capable drivers (450.80 or later and 525.60 or later respectively).
However, using the most recent driver may enable additional functionality. 
More details can be found in enhanced CUDA compatibility guide.

Install via pip for CUDA 12.0:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-cuda120==1.36.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda120==1.36.0

For CUDA 11:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-cuda110==1.36.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda110==1.36.0

Or use direct download links (CUDA 12.0):

Or use direct download links (CUDA 11.0):

FFmpeg source code:

This software uses code of FFmpeg licensed under the LGPLv2.1 and its source can be downloaded here

Libsndfile source code:

https://developer.download.nvidia.com/compute/redist/nvidia-dali/libsndfile-1.2.2.tar.gz

NVIDIA/DALI v1.36.0 DALI v1.36.0 on GitHub