Merge pull request #131 from directgroup/add-recurrentvarnet

Add recurrentvarnet implementation

README.md

@@ -19,7 +19,7 @@ See [getting_started.md](getting_started.md), check out the [documentation](http
 In the [projects](projects) folder examples are given on how to train models on public datasets.
 
 ## Baselines and trained models
-
+- [Recurrent Variational Network (RecurrentVarNet)](https://arxiv.org/abs/2111.09639)
 - [Recurrent Inference Machine (RIM)](https://www.sciencedirect.com/science/article/abs/pii/S1361841518306078)
 - [End-to-end Variational Network (VarNet)](https://arxiv.org/pdf/2004.06688.pdf) 
 - [Learned Primal Dual Network (LDPNet)](https://arxiv.org/abs/1707.06474)

direct/nn/recurrentvarnet/__init__.py

@@ -0,0 +1,2 @@
+# coding=utf-8
+# Copyright (c) DIRECT Contributors

direct/nn/recurrentvarnet/config.py

@@ -0,0 +1,20 @@
+# coding=utf-8
+# Copyright (c) DIRECT Contributors
+
+from dataclasses import dataclass
+from typing import Optional, Tuple
+
+from direct.config.defaults import ModelConfig
+
+
+@dataclass
+class RecurrentVarNetConfig(ModelConfig):
+    num_steps: int = 15  # :math:`T`
+    recurrent_hidden_channels: int = 64
+    recurrent_num_layers: int = 4  # :math:`n_l`
+    no_parameter_sharing: bool = True
+    learned_initializer: bool = True
+    initializer_initialization: Optional[str] = "sense"
+    initializer_channels: Optional[Tuple[int, ...]] = (32, 32, 64, 64)  # :math:`n_d`
+    initializer_dilations: Optional[Tuple[int, ...]] = (1, 1, 2, 4)  # :math:`p`
+    initializer_multiscale: int = 1