add attribution for dilation in convolution

src/caffe/layers/base_conv_layer.cpp

@@ -105,6 +105,9 @@ void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
   for (int i = 0; i < num_spatial_axes_; ++i) {
     dilation_data[i] = (num_dilation_dims == 0) ? kDefaultDilation :
                        conv_param.dilation((num_dilation_dims == 1) ? 0 : i);
+    if (reverse_dimensions()) {
+      CHECK_EQ(dilation_data[i], 1) << "Deconvolution doesn't support dilation";
+    }
   }
   // Special case: im2col is the identity for 1x1 convolution with stride 1
   // and no padding, so flag for skipping the buffer and transformation.

src/caffe/proto/caffe.proto

@@ -518,6 +518,13 @@ message ConvolutionParameter {
   repeated uint32 pad = 3; // The padding size; defaults to 0
   repeated uint32 kernel_size = 4; // The kernel size
   repeated uint32 stride = 6; // The stride; defaults to 1
+  // Properties of dilated convolution are described in
+  // F. Yu et al, Multi-Scale Context Aggregation by Dilated Convolutions, 2015.
+  // "dilation" was first proposed and used in a trous algorithm, as decribed in
+  // Holschneider et al. A Real-Time Algorithm for Signal Analysis
+  // with the Help of the Wavelet Transform, 1987.
+  // It was called filter rarefaction in Long* and Shelhamer* and Darrell,
+  // Fully Convolutional Networks for Semantic Segmentation, 2014.
   repeated uint32 dilation = 18; // The dilation; defaults to 1
 
   // For 2D convolution only, the *_h and *_w versions may also be used to

src/caffe/solvers/adam_solver.cpp

@@ -30,7 +30,7 @@ void AdamSolver<Dtype>::ComputeUpdateValue(int param_id, Dtype rate) {
   Blob<Dtype>* val_v = this->history_[param_id + update_history_offset].get();
   Blob<Dtype>* val_t = this->temp_[param_id].get();
 
-  const int t = this->iter_ + 1;
+  const int t = this->iter_  + 1;
   const Dtype correction = std::sqrt(Dtype(1) - pow(beta2, t)) /
       (Dtype(1.) - pow(beta1, t));
   const int N = net_params[param_id]->count();