Reference implementations of popular deep learning models.
Minor fixes.
ResNet101
, ResNet152
variants of ResNet.ResNet50V2
, ResNet101V2
, ResNet152V2
variants of ResNet V2.ResNeXt50
, ResNeXt101
variants of ResNeXt.Special thanks to @taehoonlee, the primary maintainer, who makes the existence of this module possible. We also thank @toddrme2178 who contributed to this release.
This release removes the dependency on the Keras engine
submodule (which was due to the use of the get_source_inputs
utility). Now get_source_inputs
can be imported from the utils
Keras module.
For the time being, set_keras_submodules
still supports an engine
argument in order to maintain compatibility with Keras 2.2.0. After the release of Keras 2.2.1, this argument will be removed.
In preparation of Keras 2.2.1 release.
DenseNet
pretrained weight files.MobileNet
.Fixes pre-trained weights loading issue with NASNet.
Initial release.