LEAKY plugin

include/Layer.h

@@ -97,6 +97,8 @@ typedef enum {
 class Activation : public Layer {
 
 public:
+    int act_mode;
+
     Activation(Network *net, int act_mode); 
     virtual ~Activation();
     virtual layerType_t getLayerType() { return LAYER_ACTIVATION; };
@@ -104,7 +106,6 @@ public:
     virtual value_type* infer(dataDim_t &dim, value_type* srcData);
 
 protected:
-    int act_mode;
     cudnnActivationDescriptor_t activDesc;
 };
 

src/NetworkRT.cpp

@@ -4,6 +4,7 @@
 #include "NetworkRT.h"
 
 using namespace nvinfer1;
+#include "pluginsRT/ActivationLeakyRT.cpp"
 
 // Logger for info/warning/errors
 class Logger : public ILogger			
@@ -183,14 +184,21 @@ ITensor* NetworkRT::convert_layer(ITensor *input, Pooling *l) {
 ITensor* NetworkRT::convert_layer(ITensor *input, Activation *l) {
     std::cout<<"convert Activation\n";
 
+    if(l->act_mode == ACTIVATION_LEAKY) {
+        std::cout<<"New plugin LEAKY\n";
+        IPlugin *plugin = new ActivationLeakyRT();
+        IPluginLayer *lRT = networkRT->addPlugin(&input, 1, *plugin);
+        checkNULL(lRT);
+        return lRT->getOutput(0);
+    }
+
     IActivationLayer *lRT = networkRT->addActivation(*input, ActivationType::kRELU);
     checkNULL(lRT);
-
     return lRT->getOutput(0);
 }
 
 ITensor* NetworkRT::convert_layer(ITensor *input, Softmax *l) {
-    std::cout<<"convert Activation\n";
+    std::cout<<"convert softmax\n";
 
     ISoftMaxLayer *lRT = networkRT->addSoftMax(*input);
     checkNULL(lRT);

src/pluginsRT/ActivationLeakyRT.cpp

@@ -0,0 +1,58 @@
+#include<cassert>
+#include "kernels.h"
+
+class ActivationLeakyRT : public IPlugin {
+
+public:
+	ActivationLeakyRT() {
+
+
+	}
+
+	~ActivationLeakyRT(){
+
+	}
+
+	int getNbOutputs() const override {
+		return 1;
+	}
+
+	Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override {
+		return inputs[0];
+	}
+
+	void configure(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, int maxBatchSize) override {
+		size = 1;
+		for(int i=0; i<outputDims[0].nbDims; i++)
+			size *= outputDims[0].d[i];
+	}
+
+	int initialize() override {
+
+		return 0;
+	}
+
+	virtual void terminate() override {
+	}
+
+	virtual size_t getWorkspaceSize(int maxBatchSize) const override {
+		return 0;
+	}
+
+	virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
+
+		activationLEAKYForward((value_type*)reinterpret_cast<const value_type*>(inputs[0]), 
+											reinterpret_cast<value_type*>(outputs[0]), size);
+		return 0;
+	}
+
+
+	virtual size_t getSerializationSize() override {
+		return 0;
+	}
+
+	virtual void serialize(void* buffer) override {
+	}
+
+	int size;
+};

tests/mnist/test_mnist.cpp

@@ -18,7 +18,7 @@ int main() {
     tkDNN::Conv2d     l2(&net, 50, 5, 5, 1, 1, 0, 0, c1_bin);
     tkDNN::Pooling    l3(&net, 2, 2, 2, 2, tkDNN::POOLING_MAX);
     tkDNN::Dense      l4(&net, 500, d2_bin);
-    tkDNN::Activation l5(&net, CUDNN_ACTIVATION_RELU);
+    tkDNN::Activation l5(&net, tkDNN::ACTIVATION_LEAKY);
     tkDNN::Dense      l6(&net, 10, d3_bin);
     tkDNN::Softmax    l7(&net);
 

tests/yolo/yolo.cpp

@@ -83,7 +83,7 @@ int main() {
     tkDNN::Activation a23(&net, tkDNN::ACTIVATION_LEAKY);
     tkDNN::Conv2d     c24(&net, 1024, 3, 3, 1, 1, 1, 1, c24_bin, true);
     tkDNN::Activation a24(&net, tkDNN::ACTIVATION_LEAKY);
-
+/*
     tkDNN::Layer *m25_layers[1] = { &a16 };
     tkDNN::Route      m25(&net, m25_layers, 1);
     tkDNN::Conv2d     c26(&net, 64, 1, 1, 1, 1, 0, 0,   c26_bin, true);
@@ -98,7 +98,7 @@ int main() {
     tkDNN::Conv2d     c30(&net, 425, 1, 1, 1, 1, 0, 0,  c30_bin, false);
 
     tkDNN::Region     g31(&net, 80, 4, 5, 0.6f);
-
+*/
     // Load input
     value_type *data;
     value_type *input_h;
@@ -108,16 +108,16 @@ int main() {
 
     value_type *out_data, *out_data2;
 
+    tkDNN::dataDim_t dim1 = dim;
     std::cout<<"CUDNN inference:\n"; {
-        dim.print(); //print initial dimension  
+        dim1.print(); //print initial dimension  
         TIMER_START
-        out_data = net.infer(dim, data);    
+        out_data = net.infer(dim1, data);    
         TIMER_STOP
-        dim.print();   
+        dim1.print();   
     }
  
-    tkDNN::dataDim_t dim2(1, 3, 608, 608, 1);
+    tkDNN::dataDim_t dim2 = dim;
-
     std::cout<<"TENSORRT inference:\n"; {
         dim2.print();
         TIMER_START
@@ -127,7 +127,7 @@ int main() {
     }
 
     std::cout<<"\n======= CHECK RESULT =======\n";
-    std::cout<<"Diffs: "<<checkResult(dim.tot(), out_data, out_data2)<<"\n";
+    std::cout<<"Diffs: "<<checkResult(net.getOutputDim().tot(), out_data, out_data2)<<"\n";
 
     return 0;
 }