Some advanced keras usage, like self-defined layer...
1.self-defined layer
Self-defined Layer by keras.layers.core.Lambda or keras.engine.topology.Layer.
2.seq_to_seq_addition
Use the many-to-many structure in RNN to complete the addition of two numbers. Encode numbers in a new encoding way, and use the seq_to_seq model to solve it.
3.unpooling
Unpooling is a popular method in the semantic segmentation task to segment not so-clearly border between objects. It works like below.
4.crf_text
CRF belongs to the sequence family, considering "neighboring" smaples, as well as the full context, using it keras_contrib.layers.CRF
5.crf_img_unary
6.crf_img_bilateral
CRF belongs to the sequence family, considering "neighboring" smaples, as well as the full context, using it pydensecrf