• Discuss in Slack (invited link)

• Docs

  • installation
  • layers
  • the links in https://github.com/tensorlayer/awesome-tensorlayer
  • https://tensorlayer.readthedocs.io/en/2.0.0/user/examples.html
  • https://tensorlayer.readthedocs.io/en/latest/modules/layers.html#batch-normalization batchnorm1d 2d 3d should be separated Rundi
  • others

• Layer

  • Activation act supports string inputs e.g., Dense(100, act='relu') Ruihai
  • RNN cell supports string inputs e.g., cell_fn='LSTM')
  • Remove as_layer()
  • Quantized layers should be refactored
  • tl.optimizers should be refactored
  • Optimize the `build' of core layer
  • InstanceNorm 1d 2d 3d should be implemented Yingda
  • automatically deal with in_channels:
    super(MyLayer, self).__init__(name, in_channel=in_channel)
  • refactor QuanConv2dWithBN and etc
  • Remove below codes in layers

        self.build()
        self._built = True

• Model

  • Fix model core _construct_graph BFS. @ChrisWu1997
  • Add callback features for specific layer output. @ChrisWu1997

• Visualize

  • tl.vis.save_images if input is a tensor, run x.numpy() automatically.

• Examples (Tutorial)

• text_generation

• text_ptb

• database [WIP] @ChrisWu1997

• distributed_training

• tutorial_binarynet_cifar10_tfrecord.py

• tutorial_binarynet_mnist_cnn.py

• more in quantized_net

• Initializer

  • ours vs Keras's

• Other Examples

  • https://github.com/tensorlayer/srgan
  • https://github.com/tensorlayer/openpose-plus
  • https://github.com/tensorlayer/adaptive-style-transfer
  • https://github.com/tensorlayer/seq2seq-chatbot
  • more in https://github.com/tensorlayer/awesome-tensorlayer

• Next step

  • PointCNN
  • RL - ray/rllib