mmr/tkDNN
Archived
1
0
Fork 0
Code Issues 46 Pull Requests 12 Releases 4 Activity

Add Mobilenet2SSDLite test

Browse Source 
The new test works both with TensorRT and cuDNN. Preprocessing and
Postprocessing are missing. Add ClippedReLU (for ReLU6), groups for
Conv2d, additional bias for convolution.

Other minors:
-move the timer in the detector to measure all the
processing time for a given frame (both centernet and yolo);
-add int8 flag.

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
Davide Sapienza <sapienza.dav@gmail.com>
This commit is contained in:
xavier
2020-02-21 10:45:46 +01:00
parent 40a4a55cd0
commit 808a84131c
17 changed files with 649 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/tkDNN/Layer.h
+32 -20
View File
@@ -14,6 +14,8 @@ enum layerType_t {
LAYER_DECONV2D,
LAYER_DEFORMCONV2D,
LAYER_ACTIVATION,
LAYER_ACTIVATION_CRELU,
LAYER_ACTIVATION_LEAKY,
LAYER_FLATTEN,
LAYER_MULADD,
LAYER_POOLING,
@@ -52,22 +54,24 @@ public:
std::string getLayerName() {
layerType_t type = getLayerType();
switch(type) {
case LAYER_DENSE: return "Dense";
case LAYER_CONV2D: return "Conv2d";
case LAYER_DECONV2D: return "DeConv2d";
case LAYER_DEFORMCONV2D:return "DeformConv2d";
case LAYER_ACTIVATION: return "Activation";
case LAYER_FLATTEN: return "Flatten";
case LAYER_MULADD: return "MulAdd";
case LAYER_POOLING: return "Pooling";
case LAYER_SOFTMAX: return "Softmax";
case LAYER_ROUTE: return "Route";
case LAYER_REORG: return "Reorg";
case LAYER_SHORTCUT: return "Shortcut";
case LAYER_UPSAMPLE: return "Upsample";
case LAYER_REGION: return "Region";
case LAYER_YOLO: return "Yolo";
default: return "unknown";
case LAYER_DENSE: return "Dense";
case LAYER_CONV2D: return "Conv2d";
case LAYER_DECONV2D: return "DeConv2d";
case LAYER_DEFORMCONV2D: return "DeformConv2d";
case LAYER_ACTIVATION: return "Activation";
case LAYER_ACTIVATION_CRELU: return "ActivationCReLU";
case LAYER_ACTIVATION_LEAKY: return "ActivationLeaky";
case LAYER_FLATTEN: return "Flatten";
case LAYER_MULADD: return "MulAdd";
case LAYER_POOLING: return "Pooling";
case LAYER_SOFTMAX: return "Softmax";
case LAYER_ROUTE: return "Route";
case LAYER_REORG: return "Reorg";
case LAYER_SHORTCUT: return "Shortcut";
case LAYER_UPSAMPLE: return "Upsample";
case LAYER_REGION: return "Region";
case LAYER_YOLO: return "Yolo";
default: return "unknown";
}
}
@@ -145,10 +149,18 @@ class Activation : public Layer {
public:
int act_mode;
float ceiling;
Activation(Network *net, int act_mode);
Activation(Network *net, int act_mode, const float ceiling=0.0);
virtual ~Activation();
virtual layerType_t getLayerType() { return LAYER_ACTIVATION; };
virtual layerType_t getLayerType() {
if(act_mode == CUDNN_ACTIVATION_CLIPPED_RELU)
return LAYER_ACTIVATION_CRELU;
else if (act_mode == ACTIVATION_LEAKY)
return LAYER_ACTIVATION_LEAKY;
else
return LAYER_ACTIVATION;
};
virtual dnnType* infer(dataDim_t &dim, dnnType* srcData);
@@ -165,14 +177,14 @@ class Conv2d : public LayerWgs {
public:
Conv2d( Network *net, int out_ch, int kernelH, int kernelW,
int strideH, int strideW, int paddingH, int paddingW,
std::string fname_weights, bool batchnorm = false, bool deConv = false, bool final = false, int groups = 1);
std::string fname_weights, bool batchnorm = false, bool deConv = false, bool final = false, int groups = 1, bool additional_bias=false);
virtual ~Conv2d();
virtual layerType_t getLayerType() { return LAYER_CONV2D; };
virtual dnnType* infer(dataDim_t &dim, dnnType* srcData);
int kernelH, kernelW, strideH, strideW, paddingH, paddingW;
bool deConv;
bool deConv, additional_bias;
int groups;
protected:
include/tkDNN/Network.h
+1 -1
View File
@@ -59,7 +59,7 @@ public:
dataDim_t input_dim;
dataDim_t getOutputDim();
bool fp16, dla;
bool fp16, dla, int8;
bool dontLoadWeights;
};
include/tkDNN/NetworkRT.h
+1
View File
@@ -24,6 +24,7 @@ template<typename T> T readBUF(const char*& buffer)
using namespace nvinfer1;
#include "pluginsRT/ActivationLeakyRT.h"
#include "pluginsRT/ActivationReLUCeilingRT.h"
#include "pluginsRT/ReorgRT.h"
#include "pluginsRT/RegionRT.h"
//#include "pluginsRT/RouteRT.h"
include/tkDNN/kernels.h
+1
View File
@@ -5,6 +5,7 @@
void activationELUForward(dnnType* srcData, dnnType* dstData, int size, cudaStream_t stream = cudaStream_t(0));
void activationLEAKYForward(dnnType* srcData, dnnType* dstData, int size, cudaStream_t stream = cudaStream_t(0));
void activationReLUCeilingForward(dnnType* srcData, dnnType* dstData, int size, const float ceiling, cudaStream_t stream= cudaStream_t(0));
void activationLOGISTICForward(dnnType* srcData, dnnType* dstData, int size, cudaStream_t stream = cudaStream_t(0));
void activationSIGMOIDForward(dnnType* srcData, dnnType* dstData, int size, cudaStream_t stream = cudaStream_t(0));
include/tkDNN/pluginsRT/ActivationReLUCeilingRT.h
+62
View File
@@ -0,0 +1,62 @@
#include<cassert>
#include "../kernels.h"
class ActivationReLUCeiling : public IPlugin {
public:
ActivationReLUCeiling(const float ceiling) {
this->ceiling = ceiling;
}
~ActivationReLUCeiling(){
}
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 {
activationReLUCeilingForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]),
reinterpret_cast<dnnType*>(outputs[0]), size, ceiling, stream);
return 0;
}
virtual size_t getSerializationSize() override {
return 1*sizeof(int) + 1*sizeof(float);
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
tk::dnn::writeBUF(buf, ceiling);
tk::dnn::writeBUF(buf, size);
}
int size;
float ceiling;
};