Skip to content

TensorLayer 2.2.0
Compare
Choose a tag to compare
@luomai luomai released this 13 Sep 23:11
· 277 commits to master since this release
v2.2.0
ca882d6
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

TensorLayer 2.2.0 is a maintenance release.
It contains numerous API improvement and bug fixes.
This release is compatible with TensorFlow 2 RC1.

Added

  • Support nested layer customization (#PR 1015)
  • Support string dtype in InputLayer (#PR 1017)
  • Support Dynamic RNN in RNN (#PR 1023)
  • Add ResNet50 static model (#PR 1030)
  • Add performance test code for static models (#PR 1041)

Changed

  • SpatialTransform2dAffine auto in_channels
  • support TensorFlow 2.0.0-rc1
  • Update model weights property, now returns its copy (#PR 1010)

Fixed

  • RNN updates: remove warnings, fix if seq_len=0, unitest (#PR 1033)
  • BN updates: fix BatchNorm1d for 2D data, refactored (#PR 1040)

Dependencies Update

Deprecated

Fixed

  • Fix tf.models.Model._construct_graph for list of outputs, e.g. STN case (PR #1010)
  • Enable better in_channels exception raise. (PR #1015)
  • Set allow_pickle=True in np.load() (#PR 1021)
  • Remove private_method decorator (#PR 1025)
  • Copy original model's trainable_weights and nontrainable_weights when initializing ModelLayer (#PR 1026)
  • Copy original model's trainable_weights and nontrainable_weights when initializing LayerList (#PR 1029)
  • Remove redundant parts in model.all_layers (#PR 1029)
  • Replace tf.image.resize_image_with_crop_or_pad with tf.image.resize_with_crop_or_pad (#PR 1032)
  • Fix a bug in ResNet50 static model (#PR 1041)

Removed

Security

Contributors

Assets 2
Loading