Skip to content

DALI v0.15.0

Pre-release
Pre-release
Compare
Choose a tag to compare
@klecki klecki released this 28 Oct 15:31
v0.15.0
6774406

Bug fixes

  • Fix Transpose operator when data shape with dimension of size 1 (#1244)
  • Fix DALI_Extra clone (#1276)
  • Fix conda check in DALI TF installation script (#1284)
  • Fix problems with seeking when stream start_time is != 0. (#1287)
  • Fix TypeTable initialization (#1321)
  • Fix CropMirrorNormalize compilation with GCC 8 (#1320)
  • Suppress warning when FileReader encounters dot and dot-dot entries (#1318)
  • Fix the wrong usage of find_library when searching for FFmpeg libs (#1317)
  • Fix last_batch_padded docs (#1314)
  • Fix pytorch download url (#1334)
  • Undo pytorch download changes (#1353)
  • Fix DALI TF plugin CXX11 ABI issue (#1361)
  • Add torch dependency to TL1_separate_executor (#1373)
  • Fix DALI TF installation for TF 2.0 (#1386)
  • Relax check for libnvidia-opticalflow is test script. (#1381)

Improvements

  • Replace std::pair alias with actual type (#1248)
  • Add support for volumetric (i.e. 3D) crop (depth, height and width) (#1210)
  • Refactor storage type specialization for operator aguments (#1245)
  • CPU DLTensor Operator (#1233)
  • Change Outputs and SharedOuputs return type to tuple (#1243)
  • Add non_blocking option to CopyToExternalTensor (#1254)
  • Improve heuristic for variable frame rate detection (#1242)
  • Add pipeline validation (#1267)
  • Add lookup table operator (#1251)
  • make_string for arguments, which have operator<< (#1174)
  • Tensor layout (#1237)
  • Rework Support Ops to use TensorList (#1259)
  • Improve logic in DALI TF plugin installation (support conda installation use case) (#1271)
  • size_t -> int for vec, mat, box etc... (#1277)
  • ImageDecoder libtiff implementation (#1264)
  • Add check for OF support (#1278)
  • ImageDecoder libtiff implementation (types.ANY_DATA, YCbCr, ImageDims to TensorShape) (#1280)
  • Handle nchannels>3 in ImageDecoder (#1285)
  • Use alternative compiler (e.g. g++-5.4) when available (#1290)
  • Add support for UCF-101 dataset and upgrade ffmpeg version from 3.4.2 to 4.2 (#1241)
  • Add info about libtiff dependency in the documentation (#1294)
  • Check whether random row access is allowed in libtiff based decoder implementation (#1295)
  • Make cspan (#1298)
  • BrightnessContrast operator (#1188)
  • Parse number of channels in PNGImage::PeekShape (#1288)
  • Add support for decoding multiple resolution videos in the same pipeline. (#1144)
  • Conda recipe: Point to local git repository for build source, relax version dependencies and use on conda-forge for some dependencies (#1303)
  • TiffImage::PeekShapeImpl parse and return number of channels (#1304)
  • Introduce byte_io.h including byte sequence reading utils (ReadValueBE and ReadValueLE) (#1310)
  • Add parsing of number of channels in JpegImage::PeekShapeImpl (#1306)
  • Layout refactor (#1250)
  • Add CMake VERBOSE_LOGS switch (#1319)
  • Add BMP tests (#1316)
  • Make DALI_extra repo path settable from the env (#1323)
  • Linear transformation GPU kernel (#1262)
  • Use DALI_extra images in more tests (#1177)
  • Reshape op (#1327)
  • Add tf dataset (#1299)
  • Adjust QA scripts remove installing pip whl from direct links as pip will disregard the "-f" option in that case (#1328)
  • Add CropMirrorNormalize 3D support (#1326)
  • Add layout handling to Transpose operator (#1329)
  • Add shape layout input to crop window generator signature (#1340)
  • Linear Transformation kernel for CPU (#1300)
  • Rearrange docker images (#1333)
  • Provide prebuilt plugins for manylinux2010 based pip packages (#1346)
  • Add 3D case to shape layout verification in CropAttr (#1344)

Breaking API changes

  • Change Outputs and SharedOuputs return type to tuple (#1243)

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 lesser frequency, then the returned frames may be out of sync.

  • DALI TensorFlow plugin may not be compatible with TensorFlow versions 1.15.0 and/or later. If the user wants to use DALI with TensorFlow version which doesn’t have prebuilt plugin binary shipped with DALI it requires the gcc compiler that matches the one used to build TensorFlow (gcc 4.8.4 or gcc, 4.8.5 or 5.4, depending on the particular version) is present on the system.

Binary builds

Install via pip for CUDA 9:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/cuda/9.0 nvidia-dali==0.15.0
or for CUDA 10
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/cuda/10.0 nvidia-dali==0.15.0

Or use direct download links (CUDA 9.0):

Or use direct download links (CUDA 10.0):

FFmpeg source code:

  • This software uses code of FFmpeg licensed under the LGPLv2.1 and its source can be downloaded here
Assets 2
Loading