search

dmlc / keras

Deep Learning library for Python. Convnets, recurrent neural networks, and more. Runs on MXNet, Theano or TensorFlow.
http://keras.io/
Other
125 stars 34 forks source link

Fix deconv and force compile model #55

Closed kevinthesun closed 7 years ago

kevinthesun commented 7 years ago
  1. Chop output_shape to (y, x) format to pass to target_shape.
  2. Force model compiling before training, testing and predicting.
  3. Fix pool2d when dim_ordering is 'tf'.

TODO: Current mxnet deconvolution operator only accepts target_shape which is no greater than s*(i - 1) + k, where s is strides, i is input size and k is kernel size. Maybe in the future we need to add support for oversized target_shape.