bug fixes

BrainMapperAE.py

@@ -319,6 +319,9 @@ class AutoEncoder3D(nn.Module):
 
         # Decoder
 
+        parameters['input_channels'] = parameters['output_channels']
+        parameters['convolution_stride'] = parameters['transformer_blocks_stride']
+
         if equal_channels_blocks != 0:
             self.decoderBlocks = nn.ModuleList([modules.ResNetDecoderBlock3D(parameters) for i in range(equal_channels_blocks)])
 
@@ -365,9 +368,6 @@ class AutoEncoder3D(nn.Module):
 
         # Decoder
 
-        parameters['input_channels'] = parameters['output_channels']
-        parameters['convolution_stride'] = parameters['transformer_blocks_stride']
-
         for index, decoderBlock in enumerate(self.decoderBlocks):
             if index < Y_encoder_sizes_lenght:
                 X = decoderBlock.forward(X, Y_encoder_sizes[index])

settings.ini

@@ -24,7 +24,7 @@ use_last_checkpoint = False
 adam_w_flag = False
 
 [NETWORK]
-first_kernel_size = 10
+first_kernel_size = 9
 first_convolution_stride = 1
 input_channels = 1
 output_channels = 32
@@ -34,7 +34,7 @@ convolution_stride = 2
 number_of_encoder_blocks = 2
 transformer_blocks_stride = 1
 number_of_transformer_blocks = 6
-kernel_classification = 10
+kernel_classification = 9
 pool_kernel_size = 3
 pool_stride = 2
 dropout = 0

utils/hdf5_settings.ini

@@ -24,7 +24,7 @@ negative_flag = False
 outlier_flag = True
 shrinkage_flag = False
 hard_shrinkage_flag = False
-crop_flag = False
+crop_flag = True
 input_data_train = "input_data_train.h5"
 target_data_train = "target_data_train.h5"
 input_data_validation = "input_data_validation.h5"