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

Merge with master, all tests passed

Browse Source 
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
This commit is contained in:
Micaela Verucchi
2021-07-20 12:48:41 +02:00
parent 4ca69836e9 c306b36860
commit 84ff978ccb
73 changed files with 5207 additions and 305 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/tkDNN/CenternetDetection.h
+1 -1
View File
@@ -73,7 +73,7 @@ public:
CenternetDetection() {};
~CenternetDetection() {};
bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1);
bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3);
void preprocess(cv::Mat &frame, const int bi=0);
void postprocess(const int bi=0,const bool mAP=false);
};
include/tkDNN/DarknetParser.h
+4
View File
@@ -11,6 +11,7 @@ namespace tk { namespace dnn {
int channels = 3;
int batch_normalize=0;
int groups = 1;
int group_id = 0;
int filters=1;
int size_x=1;
int size_y=1;
@@ -23,7 +24,10 @@ namespace tk { namespace dnn {
int num = 1;
int pad = 0;
int coords = 4;
int nms_kind = 0;
int new_coords= 0;
float scale_xy = 1;
float nms_thresh = 0.45;
std::vector<int> layers;
std::string activation = "linear";
include/tkDNN/DetectionNN.h
+9 -7
View File
@@ -4,7 +4,10 @@
#include <iostream>
#include <signal.h>
#include <stdlib.h>
#ifdef __linux__
#include <unistd.h>
#endif
#include <mutex>
#include "utils.h"
@@ -14,7 +17,7 @@
#include "tkdnn.h"
// #define OPENCV_CUDACONTRIB //if OPENCV has been compiled with CUDA and contrib.
//#define OPENCV_CUDACONTRIB //if OPENCV has been compiled with CUDA and contrib.
#ifdef OPENCV_CUDACONTRIB
#include <opencv2/cudawarping.hpp>
@@ -76,15 +79,15 @@ class DetectionNN {
~DetectionNN(){};
/**
* Method used to inialize the class, allocate memory and compute
* Method used to initialize the class, allocate memory and compute
* needed data.
*
* @param tensor_path path to the rt file og the NN.
* @param tensor_path path to the rt file of the NN.
* @param n_classes number of classes for the given dataset.
* @param n_batches maximum number of batches to use in inference
* @return true if everything is correct, false otherwise.
*/
virtual bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1) = 0;
virtual bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3) = 0;
/**
* This method performs the whole detection of the NN.
@@ -141,16 +144,15 @@ class DetectionNN {
}
/**
* Method to draw boundixg boxes and labels on a frame.
* Method to draw bounding boxes and labels on a frame.
*
* @param frames orginal frame to draw bounding box on.
* @param frames original frame to draw bounding box on.
*/
void draw(std::vector<cv::Mat>& frames) {
tk::dnn::box b;
int x0, w, x1, y0, h, y1;
int objClass;
std::string det_class;
int baseline = 0;
float font_scale = 0.5;
int thickness = 2;
include/tkDNN/ImuOdom.h
+9 -2
View File
@@ -1,7 +1,14 @@
#include <iostream>
#include <signal.h>
#include <stdlib.h> /* srand, rand */
#ifdef __linux__
#include <unistd.h>
#elif _WIN32
#define _USE_MATH_DEFINES
#include <math.h>
#endif
#include <mutex>
#include <Eigen/Dense>
#include "utils.h"
@@ -44,7 +51,7 @@ class ImuOdom {
virtual ~ImuOdom() {}
/**
* Method used for inizialize the class
* Method used for initialize the class
*
* @return Success of the initialization
*/
@@ -141,7 +148,7 @@ class ImuOdom {
//odomPOS = odomPOS + deltaP.cast<double>(); // V2
odomROT = odomROT * q.normalized().toRotationMatrix();
// compute euler
// compute Euler
auto newEULER = odomROT.eulerAngles(0, 1, 2);
for(int i=0; i<3; i++) {
while( fabs(newEULER(i) - odomEULER(i)) > M_PI_2 ) {
include/tkDNN/Int8BatchStream.h
+4 -1
View File
@@ -11,8 +11,11 @@
#include <fstream>
#include <iomanip>
#include <signal.h>
#include <stdlib.h>
#include <stdlib.h>
#ifdef __linux__
#include <unistd.h>
#endif
#include <mutex>
#include "NvInfer.h"
include/tkDNN/Layer.h
+30 -19
View File
@@ -19,6 +19,7 @@ enum layerType_t {
LAYER_ACTIVATION_CRELU,
LAYER_ACTIVATION_LEAKY,
LAYER_ACTIVATION_MISH,
LAYER_ACTIVATION_LOGISTIC,
LAYER_FLATTEN,
LAYER_RESHAPE,
LAYER_RESIZE,
@@ -69,6 +70,7 @@ public:
case LAYER_ACTIVATION_CRELU: return "ActivationCReLU";
case LAYER_ACTIVATION_LEAKY: return "ActivationLeaky";
case LAYER_ACTIVATION_MISH: return "ActivationMish";
case LAYER_ACTIVATION_LOGISTIC: return "ActivationLogistic";
case LAYER_FLATTEN: return "Flatten";
case LAYER_RESHAPE: return "Reshape";
case LAYER_RESIZE: return "Resize";
@@ -173,7 +175,7 @@ public:
/**
Input layer (it doesnt need weigths)
Input layer (it doesn't need weights)
*/
class Input : public Layer {
@@ -209,16 +211,17 @@ public:
/**
Avaible activation functions
Available activation functions
*/
typedef enum {
ACTIVATION_ELU = 100,
ACTIVATION_LEAKY = 101,
ACTIVATION_MISH = 102
ACTIVATION_MISH = 102,
ACTIVATION_LOGISTIC = 103
} tkdnnActivationMode_t;
/**
Activation layer (it doesnt need weigths)
Activation layer (it doesn't need weights)
*/
class Activation : public Layer {
@@ -236,6 +239,8 @@ public:
return LAYER_ACTIVATION_LEAKY;
else if (act_mode == ACTIVATION_MISH)
return LAYER_ACTIVATION_MISH;
else if (act_mode == ACTIVATION_LOGISTIC)
return LAYER_ACTIVATION_LOGISTIC;
else
return LAYER_ACTIVATION;
};
@@ -276,8 +281,8 @@ public:
protected:
cudnnFilterDescriptor_t filterDesc;
cudnnConvolutionDescriptor_t convDesc;
cudnnConvolutionFwdAlgo_t algo;
cudnnConvolutionBwdDataAlgo_t bwAlgo;
cudnnConvolutionFwdAlgoPerf_t algo;
cudnnConvolutionBwdDataAlgoPerf_t bwAlgo;
cudnnTensorDescriptor_t biasTensorDesc;
void initCUDNN(bool back = false);
@@ -321,9 +326,9 @@ public:
virtual dnnType* infer(dataDim_t &dim, dnnType* srcData);
const bool bidirectional = true; /**> is the net bidir */
bool returnSeq = false; /**> if false return only the result of last timestep */
bool returnSeq = false; /**> if false return only the result of last timestamp */
int stateSize = 0; /**> number of hidden states */
int seqLen = 0; /**> number of timesteps */
int seqLen = 0; /**> number of timestamp */
int numLayers = 1; /**> number of internal layers */
protected:
@@ -370,7 +375,7 @@ public:
/**
Deformable Convolutionl 2d layer
Deformable Convolutional 2d layer
*/
class DeformConv2d : public LayerWgs {
@@ -469,7 +474,7 @@ protected:
/**
Avaible pooling functions (padding on tkDNN is not supported)
Available pooling functions (padding on tkDNN is not supported)
*/
typedef enum {
POOLING_MAX = 0,
@@ -480,7 +485,7 @@ typedef enum {
/**
Pooling layer
currenty supported only 2d pooing (also on 3d input)
currently supported only 2d pooing (also on 3d input)
*/
class Pooling : public Layer {
@@ -529,7 +534,7 @@ public:
class Route : public Layer {
public:
Route(Network *net, Layer **layers, int layers_n);
Route(Network *net, Layer **layers, int layers_n, int groups = 1, int group_id = 0);
virtual ~Route();
virtual layerType_t getLayerType() { return LAYER_ROUTE; };
@@ -539,12 +544,14 @@ public:
static const int MAX_LAYERS = 32;
Layer *layers[MAX_LAYERS]; //ids of layers to be merged
int layers_n; //number of layers
int groups;
int group_id;
};
/**
Reorg layer
Mantain same dimension but change C*H*W distribution
Maintains same dimension but change C*H*W distribution
*/
class Reorg : public Layer {
@@ -578,7 +585,7 @@ public:
/**
Upsample layer
Mantain same dimension but change C*H*W distribution
Maintains same dimension but change C*H*W distribution
*/
class Upsample : public Layer {
@@ -629,24 +636,28 @@ public:
int sort_class;
};
Yolo(Network *net, int classes, int num, std::string fname_weights,int n_masks=3, float scale_xy=1);
enum nmsKind_t {GREEDY_NMS=0, DIOU_NMS=1};
Yolo(Network *net, int classes, int num, std::string fname_weights,int n_masks=3, float scale_xy=1, double nms_thresh=0.45, nmsKind_t nsm_kind=GREEDY_NMS, int new_coords=0);
virtual ~Yolo();
virtual layerType_t getLayerType() { return LAYER_YOLO; };
int classes, num, n_masks;
int classes, num, n_masks, new_coords;
dnnType *mask_h, *mask_d; //anchors
dnnType *bias_h, *bias_d; //anchors
float scaleXY;
double nms_thresh;
nmsKind_t nsm_kind;
std::vector<std::string> classesNames;
virtual dnnType* infer(dataDim_t &dim, dnnType* srcData);
int computeDetections(Yolo::detection *dets, int &ndets, int netw, int neth, float thresh);
int computeDetections(Yolo::detection *dets, int &ndets, int netw, int neth, float thresh, int new_coords=0);
dnnType *predictions;
static const int MAX_DETECTIONS = 8192;
static const int MAX_DETECTIONS = 8192*2;
static Yolo::detection *allocateDetections(int nboxes, int classes);
static void mergeDetections(Yolo::detection *dets, int ndets, int classes);
static void mergeDetections(Yolo::detection *dets, int ndets, int classes, double nms_thresh=0.45, nmsKind_t nsm_kind=GREEDY_NMS);
};
/**
include/tkDNN/MobilenetDetection.h
+1 -1
View File
@@ -65,7 +65,7 @@ public:
MobilenetDetection() {};
~MobilenetDetection() {};
bool init(const std::string& tensor_path, const int n_classes, const int n_batches=1);
bool init(const std::string& tensor_path, const int n_classes, const int n_batches=1, const float conf_thresh=0.3);
void preprocess(cv::Mat &frame, const int bi=0);
void postprocess(const int bi=0,const bool mAP=false);
};
include/tkDNN/Network.h
+4 -4
View File
@@ -7,12 +7,12 @@
namespace tk { namespace dnn {
/**
Data rapresentation beetween layers
Data representation between layers
n = batch size
c = channels
h = heigth (lines)
h = height (lines)
w = width (rows)
l = lenght (3rd dimension)
l = length (3rd dimension)
*/
struct dataDim_t {
@@ -43,7 +43,7 @@ public:
void releaseLayers();
/**
Do inferece for every added layer
Do inference for every added layer
*/
dnnType* infer(dataDim_t &dim, dnnType* data);
include/tkDNN/NetworkRT.h
+8 -2
View File
@@ -6,6 +6,7 @@
#include "Network.h"
#include "Layer.h"
#include "NvInfer.h"
#include <memory>
namespace tk { namespace dnn {
@@ -24,11 +25,12 @@ template<typename T> T readBUF(const char*& buffer)
using namespace nvinfer1;
#include "pluginsRT/ActivationLeakyRT.h"
#include "pluginsRT/ActivationLogisticRT.h"
#include "pluginsRT/ActivationReLUCeilingRT.h"
#include "pluginsRT/ActivationMishRT.h"
#include "pluginsRT/ReorgRT.h"
#include "pluginsRT/RegionRT.h"
//#include "pluginsRT/RouteRT.h"
#include "pluginsRT/RouteRT.h"
#include "pluginsRT/ShortcutRT.h"
#include "pluginsRT/YoloRT.h"
#include "pluginsRT/UpsampleRT.h"
@@ -59,6 +61,7 @@ public:
#if NV_TENSORRT_MAJOR >= 6
nvinfer1::IBuilderConfig *configRT;
#endif
nvinfer1::ICudaEngine *engineRT;
nvinfer1::IExecutionContext *contextRT;
@@ -91,7 +94,7 @@ public:
}
/**
Do inferece
Do inference
*/
dnnType* infer(dataDim_t &dim, dnnType* data);
void enqueue(int batchSize = 1);
@@ -115,6 +118,9 @@ public:
bool serialize(const char *filename);
bool deserialize(const char *filename);
};
}}
include/tkDNN/Yolo3Detection.h
+1 -1
View File
@@ -24,7 +24,7 @@ public:
Yolo3Detection() {};
~Yolo3Detection() {};
bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1);
bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3);
void preprocess(cv::Mat &frame, const int bi=0);
void postprocess(const int bi=0,const bool mAP=false);
};
include/tkDNN/evaluation.h
+4 -4
View File
@@ -73,12 +73,12 @@ double computeMap( std::vector<Frame> &images,const int classes,
* all the recall levels are evaluated, otherwise only
* map_point recall levels are used. For COCO evaluation
* 101 points are used.
* @param map_step step used to increment IoU theshold
* @param map_step step used to increment IoU threshold
* @param map_levels number of IoU step to perform
* @param verbose is set to true, prints on screen additional info
* @param write_on_file if set to true, the results produced by this function
* are written on file
* @param net name of the considerd neural network
* @param net name of the considered neural network
*
* @return mAP IoU_tresh:IoU_tresh+map_step*map_levels (e.g. mAP 0.5:0.95 when
* map_step=0.05 and map_levels=10)
@@ -89,7 +89,7 @@ double computeMapNIoULevels(std::vector<Frame> &images,const int classes,
const int map_levels=10, const bool verbose=false,
const bool write_on_file = false, std::string net = "");
/**
* This method computes the numper of True Positive (TP), False Positive (FP),
* This method computes the number of True Positive (TP), False Positive (FP),
* False Negative (FN), precision, recall and f1-score.
* Those values are computer over all the detections, over all the classes.
*
@@ -101,7 +101,7 @@ double computeMapNIoULevels(std::vector<Frame> &images,const int classes,
* @param verbose is set to true, prints on screen additional info
* @param write_on_file if set to true, the results produced by this function
* are written on file
* @param net name of the considerd neural network
* @param net name of the considered neural network
*/
void computeTPFPFN( std::vector<Frame> &images,const int classes,
const float IoU_thresh=0.5, const float conf_thresh=0.3,
include/tkDNN/pluginsRT/ActivationLeakyRT.h
+2 -2
View File
@@ -51,9 +51,9 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
tk::dnn::writeBUF(buf, slope);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, size);
assert(buf == a + getSerializationSize());
}
int size;
include/tkDNN/pluginsRT/ActivationLogisticRT.h
+60
View File
@@ -0,0 +1,60 @@
#include<cassert>
#include "../kernels.h"
class ActivationLogisticRT : public IPlugin {
public:
ActivationLogisticRT() {
}
~ActivationLogisticRT(){
}
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 {
activationLOGISTICForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]),
reinterpret_cast<dnnType*>(outputs[0]), batchSize*size, stream);
return 0;
}
virtual size_t getSerializationSize() override {
return 1*sizeof(int);
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
tk::dnn::writeBUF(buf, size);
}
int size;
};
include/tkDNN/pluginsRT/ActivationMishRT.h
+2 -1
View File
@@ -52,8 +52,9 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, size);
assert(buf == a + getSerializationSize());
}
int size;
include/tkDNN/pluginsRT/ActivationReLUCeilingRT.h
+2 -1
View File
@@ -51,9 +51,10 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, ceiling);
tk::dnn::writeBUF(buf, size);
assert(buf = a + getSerializationSize());
}
include/tkDNN/pluginsRT/ActivationSigmoidRT.h
+2 -1
View File
@@ -52,8 +52,9 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, size);
assert(buf == a + getSerializationSize());
}
int size;
include/tkDNN/pluginsRT/DeformableConvRT.h
+3 -2
View File
@@ -89,7 +89,7 @@ public:
for(int b=0; b<batchSize; b++) {
checkCuda(cudaMemcpy(offset, output_conv + b * 3 * chunk_dim, 2*chunk_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice));
checkCuda(cudaMemcpy(mask, output_conv + b * 3 * chunk_dim + 2*chunk_dim, chunk_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice));
// kernel sigmoide
// kernel sigmoid
activationSIGMOIDForward(mask, mask, chunk_dim);
// deformable convolution
dcnV2CudaForward(stat, handle,
@@ -116,7 +116,7 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, chunk_dim);
tk::dnn::writeBUF(buf, kh);
tk::dnn::writeBUF(buf, kw);
@@ -163,6 +163,7 @@ public:
for(int i=0; i<dim_ones; i++)
tk::dnn::writeBUF(buf, aus[i]);
free(aus);
assert(buf == a + getSerializationSize());
}
cublasStatus_t stat;
include/tkDNN/pluginsRT/FlattenConcatRT.h
+2 -1
View File
@@ -65,12 +65,13 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a = buf;
tk::dnn::writeBUF(buf, c);
tk::dnn::writeBUF(buf, h);
tk::dnn::writeBUF(buf, w);
tk::dnn::writeBUF(buf, rows);
tk::dnn::writeBUF(buf, cols);
assert(buf == a + getSerializationSize());
}
int c, h, w;
include/tkDNN/pluginsRT/MaxPoolingFixedSizeRT.h
+2 -1
View File
@@ -55,7 +55,7 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, this->c);
tk::dnn::writeBUF(buf, this->h);
@@ -65,6 +65,7 @@ public:
tk::dnn::writeBUF(buf, this->stride_W);
tk::dnn::writeBUF(buf, this->winSize);
tk::dnn::writeBUF(buf, this->padding);
assert(buf == a + getSerializationSize());
}
int n, c, h, w;
include/tkDNN/pluginsRT/RegionRT.h
+2 -1
View File
@@ -73,13 +73,14 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, classes);
tk::dnn::writeBUF(buf, coords);
tk::dnn::writeBUF(buf, num);
tk::dnn::writeBUF(buf, c);
tk::dnn::writeBUF(buf, h);
tk::dnn::writeBUF(buf, w);
assert(buf == a + getSerializationSize());
}
int c, h, w;
include/tkDNN/pluginsRT/ReorgRT.h
+2 -1
View File
@@ -52,11 +52,12 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, stride);
tk::dnn::writeBUF(buf, c);
tk::dnn::writeBUF(buf, h);
tk::dnn::writeBUF(buf, w);
assert(buf == a + getSerializationSize());
}
int c, h, w, stride;
include/tkDNN/pluginsRT/ReshapeRT.h
+2 -1
View File
@@ -50,11 +50,12 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a = buf;
tk::dnn::writeBUF(buf, n);
tk::dnn::writeBUF(buf, c);
tk::dnn::writeBUF(buf, h);
tk::dnn::writeBUF(buf, w);
assert(buf == a + getSerializationSize());
}
int n, c, h, w;
include/tkDNN/pluginsRT/ResizeLayerRT.h
+2 -1
View File
@@ -52,7 +52,7 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, o_c);
tk::dnn::writeBUF(buf, o_h);
@@ -61,6 +61,7 @@ public:
tk::dnn::writeBUF(buf, i_c);
tk::dnn::writeBUF(buf, i_h);
tk::dnn::writeBUF(buf, i_w);
assert(buf == a + getSerializationSize());
}
int i_c, i_h, i_w, o_c, o_h, o_w;
include/tkDNN/pluginsRT/RouteRT.h
+21 -11
View File
@@ -8,7 +8,9 @@ class RouteRT : public IPlugin {
*/
public:
RouteRT() {
RouteRT(int groups, int group_id) {
this->groups = groups;
this->group_id = group_id;
}
~RouteRT(){
@@ -22,7 +24,7 @@ public:
Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override {
int out_c = 0;
for(int i=0; i<nbInputDims; i++) out_c += inputs[i].d[0];
return DimsCHW{out_c, inputs[0].d[1], inputs[0].d[2]};
return DimsCHW{out_c/groups, inputs[0].d[1], inputs[0].d[2]};
}
void configure(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, int maxBatchSize) override {
@@ -34,6 +36,7 @@ public:
}
h = inputDims[0].d[1];
w = inputDims[0].d[2];
c /= groups;
}
int initialize() override {
@@ -49,15 +52,18 @@ public:
}
virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
dnnType *dstData = reinterpret_cast<dnnType*>(outputs[0]);
int offset = 0;
for(int i=0; i<in; i++) {
dnnType *input = (dnnType*)reinterpret_cast<const dnnType*>(inputs[i]);
int in_dim = c_in[i]*h*w;
checkCuda( cudaMemcpyAsync(dstData + offset, input, in_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream) );
offset += in_dim;
for(int b=0; b<batchSize; b++) {
int offset = 0;
for(int i=0; i<in; i++) {
dnnType *input = (dnnType*)reinterpret_cast<const dnnType*>(inputs[i]);
int in_dim = c_in[i]*h*w;
int part_in_dim = in_dim / this->groups;
checkCuda( cudaMemcpyAsync(dstData + b*c*w*h + offset, input + b*c*w*h*groups + this->group_id*part_in_dim, part_in_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream) );
offset += part_in_dim;
}
}
return 0;
@@ -65,11 +71,13 @@ public:
virtual size_t getSerializationSize() override {
return (4+MAX_INPUTS)*sizeof(int);
return (6+MAX_INPUTS)*sizeof(int);
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, groups);
tk::dnn::writeBUF(buf, group_id);
tk::dnn::writeBUF(buf, in);
for(int i=0; i<MAX_INPUTS; i++)
tk::dnn::writeBUF(buf, c_in[i]);
@@ -77,10 +85,12 @@ public:
tk::dnn::writeBUF(buf, c);
tk::dnn::writeBUF(buf, h);
tk::dnn::writeBUF(buf, w);
assert(buf == a + getSerializationSize());
}
static const int MAX_INPUTS = 4;
int in;
int c_in[MAX_INPUTS];
int c, h, w;
int groups, group_id;
};
include/tkDNN/pluginsRT/ShortcutRT.h
+3 -2
View File
@@ -59,7 +59,7 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, bc);
tk::dnn::writeBUF(buf, bh);
tk::dnn::writeBUF(buf, bw);
@@ -67,7 +67,8 @@ public:
tk::dnn::writeBUF(buf, c);
tk::dnn::writeBUF(buf, h);
tk::dnn::writeBUF(buf, w);
assert(buf == a + getSerializationSize());
}
int c, h, w;
include/tkDNN/pluginsRT/UpsampleRT.h
+2 -1
View File
@@ -54,11 +54,12 @@ public:
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, stride);
tk::dnn::writeBUF(buf, c);
tk::dnn::writeBUF(buf, h);
tk::dnn::writeBUF(buf, w);
assert(buf == a + getSerializationSize());
}
int c, h, w, stride;
include/tkDNN/pluginsRT/YoloRT.h
+44 -24
View File
@@ -8,12 +8,15 @@ class YoloRT : public IPlugin {
public:
YoloRT(int classes, int num, tk::dnn::Yolo *yolo = nullptr, int n_masks=3, float scale_xy=1) {
YoloRT(int classes, int num, tk::dnn::Yolo *yolo = nullptr, int n_masks=3, float scale_xy=1, float nms_thresh=0.45, int nms_kind=0, int new_coords=0) {
this->classes = classes;
this->num = num;
this->n_masks = n_masks;
this->scaleXY = scale_xy;
this->nms_thresh = nms_thresh;
this->nms_kind = nms_kind;
this->new_coords = new_coords;
mask = new dnnType[n_masks];
bias = new dnnType[num*n_masks*2];
@@ -61,17 +64,23 @@ public:
checkCuda( cudaMemcpyAsync(dstData, srcData, batchSize*c*h*w*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream));
for (int b = 0; b < batchSize; ++b){
for(int n = 0; n < n_masks; ++n){
int index = entry_index(b, n*w*h, 0);
activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream);
if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
index = entry_index(b, n*w*h, 4);
activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*w*h, stream);
}
}
for (int b = 0; b < batchSize; ++b){
for(int n = 0; n < n_masks; ++n){
int index = entry_index(b, n*w*h, 0);
if (new_coords == 1){
if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
}
else{
activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream); //x,y
if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
index = entry_index(b, n*w*h, 4);
activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*w*h, stream);
}
}
}
//std::cout<<"YOLO END\n";
return 0;
@@ -79,22 +88,29 @@ public:
virtual size_t getSerializationSize() override {
return 6*sizeof(int) + sizeof(float)+ n_masks*sizeof(dnnType) + num*n_masks*2*sizeof(dnnType) + YOLORT_CLASSNAME_W*classes*sizeof(char);
return 8*sizeof(int) + 2*sizeof(float)+ n_masks*sizeof(dnnType) + num*n_masks*2*sizeof(dnnType) + YOLORT_CLASSNAME_W*classes*sizeof(char);
}
virtual void serialize(void* buffer) override {
char *buf = reinterpret_cast<char*>(buffer);
tk::dnn::writeBUF(buf, classes);
tk::dnn::writeBUF(buf, num);
tk::dnn::writeBUF(buf, n_masks);
tk::dnn::writeBUF(buf, c);
tk::dnn::writeBUF(buf, h);
tk::dnn::writeBUF(buf, w);
tk::dnn::writeBUF(buf, scaleXY);
for(int i=0; i<n_masks; i++)
tk::dnn::writeBUF(buf, mask[i]);
for(int i=0; i<n_masks*2*num; i++)
tk::dnn::writeBUF(buf, bias[i]);
char *buf = reinterpret_cast<char*>(buffer),*a=buf;
tk::dnn::writeBUF(buf, classes); //std::cout << "Classes :" << classes << std::endl;
tk::dnn::writeBUF(buf, num); //std::cout << "Num : " << num << std::endl;
tk::dnn::writeBUF(buf, n_masks); //std::cout << "N_Masks" << n_masks << std::endl;
tk::dnn::writeBUF(buf, scaleXY); //std::cout << "ScaleXY :" << scaleXY << std::endl;
tk::dnn::writeBUF(buf, nms_thresh); //std::cout << "nms_thresh :" << nms_thresh << std::endl;
tk::dnn::writeBUF(buf, nms_kind); //std::cout << "nms_kind : " << nms_kind << std::endl;
tk::dnn::writeBUF(buf, new_coords); //std::cout << "new_coords : " << new_coords << std::endl;
tk::dnn::writeBUF(buf, c); //std::cout << "C : " << c << std::endl;
tk::dnn::writeBUF(buf, h); //std::cout << "H : " << h << std::endl;
tk::dnn::writeBUF(buf, w); //std::cout << "C : " << c << std::endl;
for (int i = 0; i < n_masks; i++)
{
tk::dnn::writeBUF(buf, mask[i]); //std::cout << "mask[i] : " << mask[i] << std::endl;
}
for (int i = 0; i < n_masks * 2 * num; i++)
{
tk::dnn::writeBUF(buf, bias[i]); //std::cout << "bias[i] : " << bias[i] << std::endl;
}
// save classes names
for(int i=0; i<classes; i++) {
@@ -104,11 +120,15 @@ public:
tk::dnn::writeBUF(buf, tmp[j]);
}
}
assert(buf == a + getSerializationSize());
}
int c, h, w;
int classes, num, n_masks;
float scaleXY;
float nms_thresh;
int nms_kind;
int new_coords;
std::vector<std::string> classesNames;
dnnType *mask;
include/tkDNN/test.h
+5 -4
View File
@@ -20,7 +20,7 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
}
if(output_bins.size() != outputs.size()) {
std::cout<<output_bins.size()<<" "<<outputs.size()<<"\n";
FatalError("outputs size missmatch");
FatalError("outputs size mismatch");
}
// Load input
@@ -29,7 +29,8 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
readBinaryFile(input_bins[0], net->input_dim.tot(), &input_h, &data);
// outputs
dnnType *cudnn_out[outputs.size()], *rt_out[outputs.size()];
//dnnType *cudnn_out[outputs.size()], *rt_out[outputs.size()];
std::vector<dnnType *> cudnn_out,rt_out;
tk::dnn::dataDim_t dim1 = net->input_dim; //input dim
printCenteredTitle(" CUDNN inference ", '=', 30); {
@@ -39,7 +40,7 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
TKDNN_TSTOP
dim1.print();
}
for(int i=0; i<outputs.size(); i++) cudnn_out[i] = outputs[i]->dstData;
for(int i=0; i<outputs.size(); i++) cudnn_out.push_back(outputs[i]->dstData);
if(netRT != nullptr) {
tk::dnn::dataDim_t dim2 = net->input_dim;
@@ -50,7 +51,7 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
TKDNN_TSTOP
dim2.print();
}
for(int i=0; i<outputs.size(); i++) rt_out[i] = (dnnType*)netRT->buffersRT[i+1];
for(int i=0; i<outputs.size(); i++) rt_out.push_back((dnnType*)netRT->buffersRT[i+1]);
}
int ret_cudnn = 0, ret_tensorrt = 0, ret_cudnn_tensorrt = 0;
include/tkDNN/utils.h
+13
View File
@@ -12,8 +12,12 @@
#include <cublas_v2.h>
#include <cudnn.h>
#ifdef __linux__
#include <unistd.h>
#endif
#include <ios>
#include <chrono>
#define dnnType float
@@ -39,6 +43,7 @@
#define TKDNN_VERBOSE 1
// Simple Timer
#ifdef __linux__
#define TKDNN_TSTART timespec start, end; \
clock_gettime(CLOCK_MONOTONIC, &start);
@@ -48,6 +53,14 @@
if(show) std::cout<<col<<"Time:"<<std::setw(16)<<t_ns<<" ms\n"<<COL_END;
#define TKDNN_TSTOP TKDNN_TSTOP_C(COL_CYANB, TKDNN_VERBOSE)
#elif _WIN32
#define TKDNN_TSTART auto start = std::chrono::high_resolution_clock::now();
#define TKDNN_TSTOP auto stop = std::chrono::high_resolution_clock::now(); \
std::chrono::duration<double> duration = stop -start; \
auto time_ms = std::chrono::duration_cast<std::chrono::milliseconds>(duration);\
double t_ns = time_ms.count();
#endif
/********************************************************
* Prints the error message, and exits