Skip to content

Latest commit

 

History

History
60 lines (40 loc) · 1.86 KB

CHANGELOG.md

File metadata and controls

60 lines (40 loc) · 1.86 KB
Raw

Dec 25, 2016

  • Supported dilated convolution
  • Memory optimization is introduced to save memory during training and testing. Wiki
  • BatchReductionLayer supports reduction on an arbitrary axis with cuda implementation.
  • Other small fixes.

Apr 27, 2016

Features:

  • Supported cuDNN v5
  • Use the cuDNN's BatchNormalization implementation as the default engine for BN layer
  • BN layer will now store running variance in its fourth blob.
  • the script python/bn_convert_style.py is added to help converting the bn style forth and back.

Dec 23, 2015

Features:

  • Implemented a planning algorithm to globally optimize the cudnn workspace consumption and speed trade-off.
  • Now richness parameter specifies the total memory in MBs available to cudnn for convolution workspaces.
  • Now the framework will try to find the best convolution algorithm combinations under memory limit.

Dec 17, 2015

Features:

  • cuDNN v4 support
  • 20% overall speed gain with faster convolution and batch normalization
  • the native batch normalization is changed to comply with cuDNN. Use the script python/bn_var_to_inv_std.py to upgrade your models.

Nov 22, 2015

Features:

  • python layer can expose a prefetch() method, which will be run in parallel with network processing.

Oct 13, 2015

Features:

  • Improved cuDNN wrapper to use less GPU memory.
  • Now there is a new parameter richness which controls the limit of workspace for cuDNN.

Sep 30, 2015

Features:

  • Support for cuDNN v3.

Sep. 7, 2015

Features:

  • New mechanism for parallel comminucation reduced parallel overhead.
  • Batch normalization, courtesy of @Cysu.

Jul, 2015

Features:

  • Action recognition tools, scripts, and examples.
  • Basic parallel training support
  • Various extra data augmentations