Merge pull request #1183 from dkurt:fix_allocation_issues

Author: vpisarev

modules/dnn/src/layers/blank_layer.cpp

@@ -49,18 +49,16 @@ class BlankLayerImpl : public BlankLayer
 public:
     BlankLayerImpl(const LayerParams&) {}
 
-    void allocate(const std::vector<Mat*> &inputs, std::vector<Mat> &outputs)
+    bool getMemoryShapes(const std::vector<MatShape> &inputs,
+                         const int requiredOutputs,
+                         std::vector<MatShape> &outputs,
+                         std::vector<MatShape> &internals) const
     {
-        outputs.resize(inputs.size());
-        for (size_t i = 0; i < inputs.size(); i++)
-            outputs[i] = *inputs[i];
+        Layer::getMemoryShapes(inputs, requiredOutputs, outputs, internals);
+        return true;
     }
 
-    void forward(std::vector<Mat*> &inputs, std::vector<Mat> &outputs, std::vector<Mat> &internals)
-    {
-        for (size_t i = 0; i < inputs.size(); i++)
-            outputs[i] = *inputs[i];
-    }
+    void forward(std::vector<Mat*> &inputs, std::vector<Mat> &outputs, std::vector<Mat> &internals) {}
 };
 
 Ptr<BlankLayer> BlankLayer::create(const LayerParams& params)

modules/dnn/src/layers/elementwise_layers.cpp

@@ -37,9 +37,9 @@ class ElementWiseLayer : public Func::Layer
     ElementWiseLayer(bool run_parallel_=false, const Func &f=Func()) : func(f), run_parallel(run_parallel_) {}
 
     bool getMemoryShapes(const std::vector<MatShape> &inputs,
-                                         const int requiredOutputs,
-                                         std::vector<MatShape> &outputs,
-                                         std::vector<MatShape> &internals) const
+                         const int requiredOutputs,
+                         std::vector<MatShape> &outputs,
+                         std::vector<MatShape> &internals) const
     {
         Layer::getMemoryShapes(inputs, requiredOutputs, outputs, internals);
         return true;

modules/dnn/src/layers/pooling_layer.cpp

@@ -203,7 +203,13 @@ class PoolingLayerImpl : public PoolingLayer
         CV_Assert(inputs.size() != 0);
         Size in(inputs[0][3], inputs[0][2]), out;
 
-        if (padMode.empty()) {
+        if (globalPooling)
+        {
+            out.height = 1;
+            out.width = 1;
+        }
+        else if (padMode.empty())
+        {
             //Yeah, something strange Caffe scheme-)
             out.height = static_cast<int>(ceil(static_cast<float>(in.height + 2 * pad.height -
                                                                   kernel.height) / stride.height)) + 1;

modules/dnn/src/layers/reshape_layer.cpp

@@ -149,6 +149,7 @@ class ReshapeLayerImpl : public ReshapeLayer
             outputs.push_back(MatShape());
             computeShapeByReshapeMask(inputs[i], newShapeDesc, newShapeRange, outputs.back());
         }
+        internals = outputs;
 
         return true;
     }
@@ -160,23 +161,28 @@ class ReshapeLayerImpl : public ReshapeLayer
         Mat srcBlob = *inputs[0];
         int dims = srcBlob.dims;
         MatShape inputShape = shape(srcBlob), outShape = shape(outputs[0]);
-        bool channelsReduced = dims > (int)outShape.size() ||
-                (dims == 4 && inputShape[1] > outShape[1]);
-        performReordering = enableReordering && dims == 4 && channelsReduced;
+
+        // input.total() == output.total(). So if reordering is require,
+        // one of the sizes will be are not equal.
+        // Example where reordering is require: from 1x128x4x4 to 1x2048
+        // Example where reordering is NOT require: from 1x1024x1x1 to 1x1024.
+        bool reorderingRequire = false;
+        const int minDims = min(dims, (int)outShape.size());
+        for (int i = 0; !reorderingRequire && i < minDims; ++i)
+            reorderingRequire = inputShape[i] != outShape[i];
+        performReordering = enableReordering && reorderingRequire;
     }
 
     void forward(std::vector<Mat*> &inputs, std::vector<Mat> &outputs, std::vector<Mat> &internals)
     {
         for (size_t i = 0; i < inputs.size(); i++)
         {
             Mat srcBlob = *inputs[i];
-            MatShape inputShape = shape(srcBlob), outShape = shape(outputs[i]);
+            MatShape inputShape = shape(srcBlob);
 
             if (performReordering)
             {
-                Mat reordered_blob(inputShape, srcBlob.type());
-
-                float *dstData = reordered_blob.ptr<float>();
+                float *dstData = internals[i].ptr<float>();
                 const float *srcData = srcBlob.ptr<float>();
 
                 int num = inputShape[0], channels = inputShape[1], height = inputShape[2], width = inputShape[3];
@@ -196,8 +202,7 @@ class ReshapeLayerImpl : public ReshapeLayer
                         }
                     }
                 }
-
-                outputs[i] = reordered_blob.reshape(1, outShape);
+                internals[i].copyTo(outputs[i]);
             }
         }
     }