Skip to content

DALI v0.14.0

Pre-release
Pre-release
Compare
Choose a tag to compare
@klecki klecki released this 30 Sep 17:37

Bug fixes

  • Fix fp16 bug from #1129 and add fp16 test case (#1160)
  • Fix framework iterators behavior when iter_setup raises StopIteration (#1136)
  • Fix nvjpeg legacy API (#1179)
  • Attempt different driver urls in setup_test_common.sh (#1193)
  • fix nightly bug in video reader (#1194)
  • Fix conversions to int64 / uint64. (#1205)
  • Attempt to fix issue with tf plugin install and gcc 4.8 (#1214)
  • Fix PyTorch spelling (#1230)

Improvements

  • BrightnessContrast CUDA kernels (#1142)
  • Adjust Operator::Run to take reference instead of pointer (#1168)
  • Add a STYLE_GUIDE for DALI, adjust Kernel example (#1167)
  • Extend external source operator capacity (#1127)
  • Make Deallocate public API (#1182)
  • Remove .cpu function (#1181)
  • Allow stream() to be called for every Workspace (#1178)
  • Improve error messages for file_list arg problems in FileReader (#1184)
  • Add multi gpu python notebook (#1186)
  • HSV Kernel for CPU (#1187)
  • Adjust CropMirrorNormalize to Setup API (#1140)
  • Expose tensor as dlpack (#1154)
  • Add const noexcept qualifiers to IsContiguous. (#1211)
  • ROI utils (#1189)
  • Add qa test for multi gpu example (#1202)
  • Add support for 3d shapes in crop window (#1207)
  • DALI for aarch64-QNX platform (#522)
  • Unified naming for float16 type. (#1212)
  • Add types to DALIDataType that were missing (#1213)
  • CPU warp, with tests. (#1159)
  • Conda Recipe for DALI (#1156)
  • Update file reader doc (#1222)
  • Track DALI_extra version in DALI (#1229)
  • Add Shapes operator returning sample shapes. (#1223)
  • New Warp operator (#1153)

Breaking API changes

  • Remove .cpu function (#1181)
  • Adjust Operator::Run to take reference instead of pointer (#1168)
  • Extend external source operator capacity (#1127) - it now requires input to be set for every iteration
  • Unified naming for float16 type. (#1212)

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"
  • 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.

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==0.14.0
or for CUDA 10
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/cuda/10.0 nvidia-dali==0.14.0

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