From eb384fbc9b5cfbbd18ea6fc9122f315fd0072f64 Mon Sep 17 00:00:00 2001
From: Rafaela Castro <rafaela00castro@gmail.com>
Date: Tue, 19 May 2020 02:47:40 +0000
Subject: [PATCH] Fix model name

---
 model/decoder.py | 65 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 65 insertions(+)
 create mode 100644 model/decoder.py

diff --git a/model/decoder.py b/model/decoder.py
new file mode 100644
index 0000000..2b2f848
--- /dev/null
+++ b/model/decoder.py
@@ -0,0 +1,65 @@
+import torch 
+import torch.nn as nn
+import torch.nn.functional as F
+import time as tm        
+        
+class DecoderSTCNN(nn.Module):
+    
+    def __init__(self, layer_size, kernel_size, initial_filter_size, channels, dropout_rate, upsample=False):
+        super(DecoderSTCNN, self).__init__()
+        self.padding = kernel_size - 1
+        self.upsample = upsample
+        self.dropout_rate = dropout_rate
+        self.conv_layers = nn.ModuleList()
+        self.relu_layers = nn.ModuleList()
+        self.batch_layers = nn.ModuleList()
+        self.dropout_layers = nn.ModuleList()
+
+        temporal_kernel_size =  [kernel_size, 1, 1]
+        temporal_padding =  [self.padding, 0, 0]
+        out_channels = initial_filter_size
+        in_channels = channels
+        for i in range(layer_size):
+            self.conv_layers.append(
+                nn.Conv3d(in_channels=in_channels, out_channels=out_channels, 
+                          kernel_size=temporal_kernel_size, padding=temporal_padding, bias=False)
+            )
+            self.relu_layers.append(nn.ReLU())
+            self.batch_layers.append(nn.BatchNorm3d(out_channels))
+            self.dropout_layers.append(nn.Dropout(dropout_rate))
+            in_channels = out_channels
+            
+        self.upsample_conv = nn.ConvTranspose3d(out_channels, out_channels, kernel_size=temporal_kernel_size, 
+                                                stride=[3,1,1], padding=[1,0,0])
+        padding_final = [kernel_size // 2, 0, 0]
+        self.conv_final = nn.Conv3d(in_channels=out_channels, out_channels=1, kernel_size=temporal_kernel_size, 
+                                    padding=padding_final, bias=True)
+        
+        
+    def learning_with_dropout(self, x):
+        for conv, relu, batch, drop in zip(self.conv_layers, self.relu_layers, 
+                                           self.batch_layers, self.dropout_layers):
+            x = conv(x)[:,:,:-self.padding,:,:]
+            x = drop(relu(batch(x)))
+            
+        return x
+    
+    def learning_without_dropout(self, x):
+        for conv, relu, batch in zip(self.conv_layers, self.relu_layers, self.batch_layers):
+            x = conv(x)[:,:,:-self.padding,:,:]
+            x = relu(batch(x))
+            
+        return x
+        
+    def forward(self, input_):
+        if self.dropout_rate > 0.:
+            output = self.learning_with_dropout(input_)
+        else:
+            output = self.learning_without_dropout(input_)
+        if (self.upsample):
+            output_size = torch.randn(input_.shape[0],1, input_.shape[2] + 10, 
+                                      input_.shape[3], input_.shape[4]).size()
+            output = self.upsample_conv(output, output_size=output_size)
+
+        output = self.conv_final(output)
+        return output
\ No newline at end of file