diff --git a/src/net.cpp b/src/net.cpp
index 50c2a4cd1627..c06c410fed1c 100644
--- a/src/net.cpp
+++ b/src/net.cpp
@@ -1271,9 +1271,10 @@ Layer* Net::create_custom_layer(int index)
 
 int Net::forward_layer(int layer_index, std::vector<Mat>& blob_mats, Option& opt) const
 {
+  printf("dims %d\n", blob_mats[0].dims);
     const Layer* layer = layers[layer_index];
 
-//     fprintf(stderr, "forward_layer %d %s\n", layer_index, layer->name.c_str());
+
 
     if (layer->one_blob_only)
     {
@@ -1283,7 +1284,15 @@ int Net::forward_layer(int layer_index, std::vector<Mat>& blob_mats, Option& opt
 
         if (blob_mats[bottom_blob_index].dims == 0)
         {
+          printf("needing %s \n", blobs[bottom_blob_index].name.c_str());
+          printf("forwarding_layer %d %s\n", blobs[bottom_blob_index].producer,
+                  layers[blobs[bottom_blob_index].producer]->name.c_str());
             int ret = forward_layer(blobs[bottom_blob_index].producer, blob_mats, opt);
+            printf("forwarded_layer %d %s\n", blobs[bottom_blob_index].producer,
+                   layers[blobs[bottom_blob_index].producer]->name.c_str());
+            printf("dims of %s : %d %d %d\n", blobs[bottom_blob_index].name.c_str(), blob_mats[bottom_blob_index].c,
+                   blob_mats[bottom_blob_index].h,
+                   blob_mats[bottom_blob_index].w);
             if (ret != 0)
                 return ret;
         }
@@ -1348,7 +1357,17 @@ int Net::forward_layer(int layer_index, std::vector<Mat>& blob_mats, Option& opt
 
             if (blob_mats[bottom_blob_index].dims == 0)
             {
-                int ret = forward_layer(blobs[bottom_blob_index].producer, blob_mats, opt);
+              printf("needing %s \n", blobs[bottom_blob_index].name.c_str());
+
+              printf("forwarding_layer %d %s\n", blobs[bottom_blob_index].producer,
+                     layers[blobs[bottom_blob_index].producer]->name.c_str());
+               int ret = forward_layer(blobs[bottom_blob_index].producer, blob_mats, opt);
+                printf("forwarded_layer %d %s\n", blobs[bottom_blob_index].producer,
+                       layers[blobs[bottom_blob_index].producer]->name.c_str());
+                printf("dims of %s:  %d %d %d\n", blobs[bottom_blob_index].name.c_str(), blob_mats[bottom_blob_index].c,
+                       blob_mats[bottom_blob_index].h,
+                       blob_mats[bottom_blob_index].w);
+
                 if (ret != 0)
                     return ret;
             }
diff --git a/tools/caffe/caffe2ncnn.cpp b/tools/caffe/caffe2ncnn.cpp
index 32f02e32350e..689cd5549c9f 100644
--- a/tools/caffe/caffe2ncnn.cpp
+++ b/tools/caffe/caffe2ncnn.cpp
@@ -390,10 +390,17 @@ int main(int argc, char** argv)
     // global definition line
     // [layer count] [blob count]
     int layer_count = proto.layer_size();
+    int layer_suppl = 0;
+    int blob_suppl = 0;
     std::set<std::string> blob_names;
     for (int i=0; i<layer_count; i++)
     {
         const caffe::LayerParameter& layer = proto.layer(i);
+        if (layer.type() == "Axpy")
+          {
+            layer_suppl+=1;
+            blob_suppl+=1;
+          }
 
         for (int j=0; j<layer.bottom_size(); j++)
         {
@@ -446,7 +453,7 @@ int main(int argc, char** argv)
             ++it;
         }
     }
-    fprintf(pp, "%lu %lu\n", layer_count + bottom_reference.size(), blob_names.size() + splitncnn_blob_count);
+    fprintf(pp, "%lu %lu\n", layer_count + bottom_reference.size() + layer_suppl, blob_names.size() + splitncnn_blob_count + blob_suppl);
 
     // populate
     blob_name_decorated.clear();
@@ -498,6 +505,25 @@ int main(int argc, char** argv)
         {
             fprintf(pp, "%-16s", "Clip");
         }
+        else if (layer.type() == "Axpy")
+        {
+          // // zero : split in order not to modiyfy x
+          // fprintf(pp, "%-16s %-16s 1 1", "Split", (layer.name() + "_split").c_str());
+          // fprintf(pp, " %s %s\n", layer.bottom(1).c_str(), (layer.bottom(1)+"_split").c_str());
+
+          //first scale layer
+          fprintf(pp, "%-16s %-16s 2 1", "Scale", (layer.name() + "_ax").c_str());
+          fprintf(pp, " %s %s %s", (layer.bottom(1) //+"_split"
+                                    ).c_str(), layer.bottom(0).c_str(),
+                  (layer.name()+"_ax").c_str());
+          fprintf(pp, " 0=-233 1=0\n");
+          //then sum layer
+          fprintf(pp, "%-16s %-16s 2 1", "Eltwise", (layer.name()+ "_xpy").c_str());
+                  fprintf(pp, " %s %s %s", layer.bottom(2).c_str(), (layer.name()+"_ax").c_str(),
+                  layer.top(0).c_str());
+                  fprintf(pp, " 0=1 -23301=2,1.0,1.0\n");
+          continue;
+        }
         else
         {
             fprintf(pp, "%-16s", layer.type().c_str());
@@ -937,7 +963,7 @@ int main(int argc, char** argv)
         }
         else if (layer.type() == "Eltwise")
         {
-            const caffe::EltwiseParameter& eltwise_param = layer.eltwise_param();
+          const caffe::EltwiseParameter& eltwise_param = layer.eltwise_param();
             int coeff_size = eltwise_param.coeff_size();
             fprintf(pp, " 0=%d", (int)eltwise_param.operation());
             fprintf(pp, " -23301=%d", coeff_size);