TracyCuiq / ResRNN

ResRNN

Implementation of a two path ResRNN architecture (Circle-RNN was invented).

This repository is an implementation of our IPMI 2017 paper:

Xue W., Nachum I.B., Pandey S., Warrington J., Leung S., Li S. (2017) Direct Estimation of Regional Wall Thicknesses via Residual Recurrent Neural Network. In: Information Processing in Medical Imaging. IPMI 2017. Lecture Notes in Computer Science, vol 10265. Springer.

https://link.springer.com/chapter/10.1007/978-3-319-59050-9_40 or https://arxiv.org/pdf/1705.09728.pdf

Two RNN modules are deployed for temporal dynamic modeling and spatial context modeling.

ON implementation of Circle-RNN

To implement Circle-RNN,the following modifications to the caffe toolbox are required:

1. add 'circle_lstm_layer.hpp' to '/caffe/include/caffe/layers', add 'circle_lstm_layer.cpp' to '/caffe/src/caffe/layers';

2. In caffe.proto, add the following line to message RecurrentParameter:

   optional uint32 depth = 6 [default = 0];

Note: LSTM units is employed in Circle-RNN.

Example of Circle-RNN

layer {
  name: "lstm1"
  type: "CircleLSTM"
  bottom: "ip1_permute"
  bottom: "clip_permute"
  top: "wt_lstm1"
  recurrent_param {
  num_output: 6
  weight_filler {
      type: "uniform"
      min: -0.05
      max: 0.05
    }
  bias_filler {
      type: "constant"
      value: 0
    }
  depth: 0
  }
}

The parameters depth indicates how many rounds the RNN network is unrolled.

• depth=0 : the number of rounds equals the number of time steps of the sequences.
• depth=1 : the RNN network is unrolled once. This is equivalent to the original RNN in caffe with LSTM unit;