NVIDIA/DALI v1.22.0 on GitHub

Key Features and Enhancements

This DALI release includes the following key features and enhancements:

Added CUDA 12.0 support (#4502).
- Reduced binary size for CUDA 12 builds.
Added CPU experimental.inputs.video operator that supports decoding video from memorybuffer across multiple iterations to reduce memory usage (#4519).
Added GPU fn.experimental.filter (convolution) operator (#4298, #4525).
Added support for decoding raw H264 and H265 streams from memory (#4480).

Fixed Issues

No major issues were fixed in this release.

Improvements

Update DALI TensorFlow examples to work with 2.11 (#4554)
Update nvCOMP to 2.5 (#4550)
Fix TL1_custom_src_pattern_build test (#4546)
Allow CPU dtype source in GPU cast_like (#4547)
Add GPU filter operator (2D, 3D) (#4525)
Remove usage of the unified memory from the remap test (#4544)
Split DALI operator tests into two jobs (#4543)
Update suppression list for sanitizer tests (#4542)
Update Boost preprocessor and rapidjson (#4538)
Update libtiff (#4531)
Fix linter errors & numpy dependency workaround (#4532)
VideoInput operator for the CPU (#4519)
Use pointer in NVDECLease. Store owner pointer in NVDECLease. (#4523)
Extract ResizeAttrBase to be reused in TensorResizeAttr (#4515)
Add GPU filter kernel (#4298)
Propagate SourceInfo (when unambiguous) from inputs to outputs. (#4518)
Limit NumPy version to pre-1.24 (#4527)
Avoid signed/unsigned comparison in clamp<S, U>. (#4524)
Update YOLO example for the latest to support the latest TensorFlow version (#4522)
Utilities and refactoring pre-VideoInput operator (#4513)
Enable CUDA 12.0 support (#4502)
Extracting InputOperator from ExternalSource (#4505)
Add expand_dims utility (#4493)
Remove Operator inheritance from VideoDecoderBase (#4508)
Extend decoding support (#4480)
Place AutoGraph as private submodule of DALI and enable tests (#4504)
Link CFITSIO library with cmake (#4487)

Bug Fixes

Add the missing installation of sanitizer to the deps image (#4521)
Fix DALI build without FFmpeg (#4534)
Replace usages of numpy.bool with bool (#4526)
Fix missing #include <optional>. (#4520)
Fix exclusion of CFITSIO test when BUILD_CFITSIO=OFF (#4510)
Don't look for duplicate arguments in parent schemas. (#4507)
Fix size argument to strncpy in cfitsio_test. Fix copyright notice. (#4509)

Breaking API changes

DALI builds for CUDA 12 dynamically link the CUDA toolkit. To use DALI, install the latest CUDA toolkit
DALI 1.21 was the last release built for CUDA 10.2.

Deprecated features

No features were deprecated in this release.

Known issues:

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==1.22.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda120==1.22.0

or for CUDA 11:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-cuda110==1.22.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda110==1.22.0

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-1.2.0.tar.gz

NVIDIA/DALI v1.22.0 DALI v1.22.0 on GitHub