Skip to content

Releases: NVIDIA/DALI

DALI v0.6.0

20 Dec 01:55
@JanuszL JanuszL
v0.6.0
c8d9d72
Compare
Choose a tag to compare
DALI v0.6.0 Pre-release
Pre-release

Bug fixes

  • Fix problem with GPU DALI operator in the TensorFlow evaluated on the CPU (#335)
  • Fix obtaining color augmentation per sample (#337)
  • Fix command line in the TF Example README (#327)
  • Fix issues reported by valgrind (#308)
  • Fixes in BBFlip and consistency in bbox format (#300)
  • Fix line endings from CRLF to LF (#315)
  • Fix for race condition on Displacement Filter Impl (#311)
  • Fixed slice coordinates calculation (#312)
  • Fix validation pipeline for accuracy in TF example (#305)
  • Fix ResizeAttr usage in Resize operator (#299)
  • Skip 0 sized images in the MxNet reader. (#303)
  • Fix tfrecord2idx compatibility for python3 (#288)
  • Fix clang build (#276)
  • TF Example: updates TF op call with the right args (#295)

Improvements

  • Add TensorFlow RN50 demo to the Sphinx documentation (#352)
  • Add rst doc for ssd pytorch example (#349)
  • Added SSD training example (#342)
  • Add base of VideoReader (#316)
  • Implement SequenceCrop Operator for CPU (#283)
  • Pytorch/MXNet plugin - use dictionary of categories (#282)
  • adding ifdef for jpeg turbo support (#341)
  • Remove NonConstRef check in cpplint.py (#340)
  • Add supported device by every operator to docs (#326)
  • Added cpu box encoder for SSD support (#325)
  • Alligns TensorFlow operator supported types with what DALI can provide (#332)
  • Sequence Reader for extracted frames (#281)
  • Add CPU operator for TensorFlow plugin with an example (#322)
  • Increase num_threads in TF RN example (#321)
  • Support for multiple labels in MXNet reader (#319)
  • Bbox crop label filtering (#320)
  • Add a wrapper for TensorFlow plugin to make pipeline serialization transparent (#310)
  • Added bounding box flipping on GPU. (#314)
  • Minimal changes for CPU CropMirrorNormalize (#257)
  • Added bounding box paste for CPU backend. (#294)
  • Add ability to return CPU TensorList as numpy array (#304)
  • Remove debug prints from async_pipelined_executor (#298)
  • Documentation Badge Added (#291)
  • TF Example: specify steps arg to tf.Estimator.evaluate for ending the evaluation (#293)
  • GPU version of RandomBBoxCrop and Slice (#269)
  • Printing the right error message in OperatorInstance init (#286)
  • Remove stat call during file discovery in the reader (#275)
  • Make libjpegturbo root dir hint preceding pkgconfig (#285)
  • Make Dali linking with static libprotobuf if possible (#284)
  • Make TensorFlow DALI operator able to return the arbitrary number of outputs (#265)

Breaking API changes

  • DALI TensorFlow operator has new API - please check examples for the reference
  • PyTorch and MXNet python iterators API has changed - please check examples for the reference

Known issues:

  • New Video reader operator requires NVIDIA VIDEO CODEC SDK support in the platform. NVIDIA GPU Cloud (NGC) optimized containers lacks this functionality in the default configuration prior to 19.01. To enable it please run the container with the ‘video’ capability enabled, ie.:
    -e "NVIDIA_DRIVER_CAPABILITIES=compute,utility,video"

Binary builds

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.6.0

Or use direct download links:

FFmpeg source code:

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

DALI v0.5.0

26 Nov 23:26
@JanuszL JanuszL
v0.5.0
235ae80
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
DALI v0.5.0 Pre-release
Pre-release

Bug fixes

  • Fixed docstring of prefetch queue depth (#263)
  • Add checking if there is any supported jpeg inside batch for batch decode (#245)
  • Add enforce for num_shards > shard_id (#246)
  • Make jupyter example fully compatible with python3 (#233)
  • Add .clang-format for Google C++ style guide (#210)
  • Update MxNet version in the README (#204)
  • Fixed race condition in AsyncPipelinedExecutor destructor (#271)

Improvements

  • Increased seed size to int64 (#252)
  • SSD support for COCO reader (#196)
  • Move PyTorch example training pipeline to the CPU (#247)
  • Add version variable to init (#250)
  • Tiff decoding (#248)
  • Object orienting image module (#222)
  • Changing Tensor::ntensor() return type (#242)
  • Type safe reader with user-provided custom-type handling (#232)
  • Add pipelined execution completion callback setter (#226)
  • Add better errors in decoders (#218)
  • Make ABI test working with installed whl (#220)
  • Added new examples to online docs (#270)
  • Added Clang to Dockerfile.deps and pass CC and CXX as arguments (#264)
  • Added example demo for ResNet with TensorFlow and DALI (#251)
  • Remove unused private field (#205)

Breaking API changes

  • Random seed type changed from INT to INT64, therefore, serialized pipelines from versions prior to 0.5 are not compatible with the current DALI version.

Binary builds

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.5.0

Or use direct download links:

Loading

DALI v0.4.1

07 Nov 20:56
@JanuszL JanuszL
v0.4.1
2a5e14c
Compare
Choose a tag to compare
DALI v0.4.1 Pre-release
Pre-release

Bug fixes

  • Fixed TF 1.11 and TF 1.12 compatibility (#237)
  • Fixed PyTorch iterator for multi-GPU (#239)

Improvements

  • Made jupyter tests executing inplace (#255)
  • Removed hardcoded pipeline length in PipelinedExecutor (#239)
  • Adjusted PyTorch example to use new nvJpeg API (#239)
  • Remove double-buffering on the MXNet side (#258)

Binary builds

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.4.1

Or use direct download links:

Loading

DALI v0.4.0

31 Oct 22:14
@JanuszL JanuszL
v0.4.0
7d3d9f8
Compare
Choose a tag to compare
DALI v0.4.0 Pre-release
Pre-release

Bug fixes

  • Fixed ability to use the same output from the support operator by CPU and GPU stage
  • Removed inconsistent-missing-override Clang warning (#197)
  • Fixed clang warnings in half.hpp and tests (#194)
  • Resolved conflicting build dirs (#189)
  • Removed the redundant imports and spaces in pytorch example (#190)
  • Fixed table in README.rst
  • Fixed reporting of the end of epoch in MXNet and pyTorch plugins (#180)
  • Fixed parsing of JPEG headers (#175)
  • Maked assigning of the classes to discovered dirs by file reader base on alphabetic order.
  • Fixed BMP size reading
  • Moved wait in multiple input sets case to the common place to guard against problem reoccurring in newly added ops
  • Removed batch_size_ from CoinFlip operator (#152)
  • Fixed corruption in MXNet reader when image is split between multiple records (#216)

Improvements

  • Added bounding box mirror operator (#188)
  • Added random crop for SSD (#176)
  • Added COCO dataset reader (#110)
  • Removed visibility of all non DALI symbols and test if ABI is clean (#191)
  • Added support for pad in MXNet plugin (#186)
  • Reduced memory usage (#195)
  • Made libprotobuf internal to DALI only (#179)
  • Added CUDA 10 based build (#178)
  • Made use epoch_size instead of hardcoded values (#174)
  • Added random paste operator (#105)
  • Added clang build (#163)
  • Added png in testing pipeline, add some of tiff routines
  • Made files to be copied after build not only when libdali is rebuild
  • Put common test code into one file
  • Upgraded OpenCV to 3.4.3 (#168)
  • Added color-twist operator (#164)
  • Changed MxNet to 1.3.0 no-beta (#183)
  • Added better sharding when number of shards does not divide the dataset size evenly (#181)
  • Updated google benchmark to v1.4.1 + several fixes (#182)
  • Added CPU versions of Crop/CropCastPermute operators (#148)
  • Added info about posting questions and problems
  • Updated PyTorch example to be alligned with the reent APEX release (#206)
  • Improved load balancing nvJPEG work (#217)
  • Updated nvJPEG to 0.2.0 version (#227)
  • Added fine grained control over output buffers in the pipeline (#212)

Binary builds

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.4.0

Or use direct download links:

Loading

DALI v0.3.0

26 Sep 17:50
@JanuszL JanuszL
v0.3.0
74d9b58
Compare
Choose a tag to compare
DALI v0.3.0 Pre-release
Pre-release

Bug fixes

  • Adjusted PyTorch Dali pipeline to be similar to MXNet example (#107)
  • Add CPU fallback for BMP images and conscious fail for GIF (#124)
  • Enable FileReader shuffling for GPU0 (#134)
  • Fix squeeze for tensor with 1 element
  • Fix segfault in MXNetReader when given bad path to index file
  • Increase timeout, parametrize Python version in Jupyter tests (#126)
  • Fix segfault in Filereader if directory does not exist.
  • Update Workspace docstrings (#111)
  • Allow pkg_config to fail in the search for JpegTurbo
  • Fixed wrong rewind in TFRecord reader (#167)

Improvements

  • Added CPU version of Resize operator (#127)
  • Added Caffe reader to TF multi reader example (#103)
  • Added filtering extensions that FileReader can read (#137)
  • Made DALI understand float16 input from python
  • Added float16 as possible output type to python
  • Added flip operator (#130)
  • Added 'at' method to TensorListGPU (#131)
  • Refactored tests (#91)
  • Shortened git SHA in the Sphinx docs to 7 chars (#108)
  • Made files to be copied during build into build_dir. (#87)
  • Added links to GTC presentation to README
  • Reduced number of pinned memory allocations (#169)

Binary builds

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.3.0

Or use direct download links:

Loading

DALI v0.2.0

27 Aug 22:40
@JanuszL JanuszL
v0.2.0
2319d24
Compare
Choose a tag to compare
DALI v0.2.0 Pre-release
Pre-release

Bug fixes

  • Avoid full construction of the pipeline during construction and fix seed support in serialized pipelines (#16)
  • Fix as_tensor not keeping the parent alive in Python (#60)
  • Fix for "invalid resource handle" in multi-gpu training
  • Fixes to PyTorch example. Need to reset DALI iterators between epochs. Putting model/loss computation back to default stream due to encountered memory access errors otherwise (#15)
  • Move example file_list to proper dir (#38)
  • Added fallback to host decoder when image is not JPEG but PNG instead (like n02105855_2933.JPEG from ImageNet) (#118)

Breaking API changes

  • The API for the Resize operator changed to match other similar operators like ResizeCropMirror.
  • The API for the TensorFlow plugin changed to allow specifying the whole shape of the tensor instead of N, H, and W separately; which enables handling both NCHW and NHWC outputs.
  • The type of labels produced by the TensorFlow plugin have changed. In DALI version 0.1.2, it was always tf.float32. In this release, a new optional parameter called label_type is introduced to the TensorFlow plugin to control the type of label. The default value for label_type is tf.int64 to better align with the label type in TFRecord.

Improvements

  • Add NVTX ranges for Operators run (#73)
  • Add a note about NGC containers in README (#78)
  • Unfused Crop operator and CropCastPermute operator (#50)
  • Make build more restrictive Werror (#71)
  • Add links to docs in README (#72)
  • Expanded TF compatibility tests
  • Add example with multiple readers pluged into TF (#58)
  • Make pkg-config optional for CMake (#59)
  • Resize refactor (#63)
  • Add type casting in Python (#54)
  • Add check that third_party git submodules are synced
  • Add fallback in cmake when .pc file is not available for libjpeg-turbo (#49)
  • Sphinx documentation (#36)
  • Fix nvJpeg include dir (#47)
  • Add private attribute naming convention to Pipeline::current_seed_ (#46)
  • Add a shape argument for the output of the TF plugin (#45)
  • Bump up libturbo-jpeg version to 1.5.3 (#44)
  • Clean up dependencies list and dependency checks (#42)
  • Switch over completely to FindProtobuf.cmake from CMake 3.9.6 (#41)
  • Update README for prerequisites (#40)
  • Add error checking for file_list format in file_loader. (#37)
  • Add test support for various versions of pyTorch (#35)
  • Add polymorphism for TF plugin outputs (#33)
  • Add tensor layout checking (#32)
  • Avoid rebuilding *.cu files during 'make install' after 'make' (#25)
  • Add CUDA 8, OpenCV 2 support and options to disable libjpeg-turbo and nvJPEG (#22)
  • Add CONTRIBUTING.md file and updated contribution section in the README.md (#20)
  • Avoid full construction of the pipeline during construction and fix seed support in serialized pipelines (#16)
  • Add int64 as label type and set it as default (#125)

Binary builds

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.2.0

Or use direct download links:

Loading

DALI v0.1.2

31 Jul 00:50
@cliffwoolley cliffwoolley
v0.1.2
d99027d
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
DALI v0.1.2 Pre-release
Pre-release

Bug fixes

  • Fix compatibility with TensorFlow 1.9 (#52)
  • Update to nvJPEG v0.1.2 to fix batched decoding when a batch contains both grayscale and color images (#79)

Improvements

  • Add Tensorflow 1.7 support (#24)
  • Better overlap when using DALI with multi-GPU in MXNet and pyTorch (#76)

Binary builds

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.1.2

Or use direct download links:

Loading

DALI v0.1.1

29 Jun 20:26
@cliffwoolley cliffwoolley
v0.1.1
6d13bc6
Compare
Choose a tag to compare
DALI v0.1.1 Pre-release
Pre-release

Bug fixes

  • #4 - Race in processing multiple input sets
  • #5 - ImportError with various shared object file dependencies not found
  • #8 - Segfault in ops.FileReader when no files found
  • #12 - Python3 incompatibility in some examples
  • #13 - Crash when importing pre-built DALI PyTorch plugin w/ pre-built PyTorch
  • Pre-built binary includes an updated NVJPEG build that fixes a race condition seen in some DALI pipelines

Improvements

  • Binary compatibility of the pre-built DALI binaries with pre-built DL frameworks is improved (#13).
    • In support of this, most dependencies are now statically linked into the pre-built binaries, and the list of symbols exported from the shared objects are significantly reduced.
    • A beneficial side effect is that CUDA 9.0 Toolkit is no longer required to be installed to use pre-built binaries; only the corresponding NVIDIA Driver is required. This for example allows compatibility with a DL framework otherwise built against CUDA 9.1 or 9.2.

Binary builds

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.1.1

Or use direct download links:

Loading

DALI v0.1.0 : Initial public release

28 Jun 22:33
@cliffwoolley cliffwoolley
v0.1.0
f4087bc
Compare
Choose a tag to compare
DALI v0.1.0 : Initial public release Pre-release
Pre-release

Binaries

Install via pip:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali==0.1.0

Or use direct download links:

Loading