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

- ImuOdom model into class

Browse Source 
- Route layer input array hard copy
- fix utils
This commit is contained in:
Francesco Gatti
2020-03-09 19:44:21 +01:00
parent cd1e1c1582
commit 6fee8d5ec5
9 changed files with 202 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.gitignore
+2 -1
View File
@@ -9,4 +9,5 @@ build/
*.tar.gz
*.weights
.idea/
*.hdf5
*.hdf5
*.pk
CMakeLists.txt
+3
View File
@@ -29,6 +29,9 @@ cuda_add_library(kernels SHARED ${tkdnn_CUSRC})
#-------------------------------------------------------------------------------
# External Libraries
#-------------------------------------------------------------------------------
find_package(Eigen3 REQUIRED)
include_directories(${EIGEN3_INCLUDE_DIR})
find_package(OpenCV REQUIRED)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DOPENCV")
include/tkDNN/ImuOdom.h
+124
View File
@@ -0,0 +1,124 @@
#include <iostream>
#include <signal.h>
#include <stdlib.h> /* srand, rand */
#include <unistd.h>
#include <mutex>
#include <Eigen/Dense>
#include "utils.h"
#include "tkdnn.h"
namespace tk { namespace dnn {
/**
*
* @author Francesco Gatti
*/
class ImuOdom {
public:
tk::dnn::Network *net = nullptr;
// Network input dim
tk::dnn::dataDim_t dim0;
tk::dnn::dataDim_t dim1;
tk::dnn::dataDim_t dim2;
// Network output dim
tk::dnn::dataDim_t odim0;
tk::dnn::dataDim_t odim1;
// input pointers
dnnType *i0_d, *i1_d, *i2_d;
// output pointers
dnnType *o0_d, *o1_d;
// output eigen CPU
Eigen::MatrixXf deltaP, deltaQ;
ImuOdom() {}
virtual ~ImuOdom() {}
/**
* Method used for inizialize the class
*
* @return Success of the initialization
*/
bool init(std::string layers_path) {
dim0 = tk::dnn::dataDim_t(1, 4, 1, 100);
dim1 = tk::dnn::dataDim_t(1, 3, 1, 100);
dim2 = tk::dnn::dataDim_t(1, 3, 1, 100);
checkCuda( cudaMalloc(&i0_d, dim0.tot()*sizeof(dnnType)) );
checkCuda( cudaMalloc(&i1_d, dim1.tot()*sizeof(dnnType)) );
checkCuda( cudaMalloc(&i2_d, dim2.tot()*sizeof(dnnType)) );
std::string c0_bin = layers_path + "/conv1d_7.bin";
std::string c1_bin = layers_path + "/conv1d_8.bin";
std::string c2_bin = layers_path + "/conv1d_9.bin";
std::string c3_bin = layers_path + "/conv1d_10.bin";
std::string c4_bin = layers_path + "/conv1d_11.bin";
std::string c5_bin = layers_path + "/conv1d_12.bin";
std::string l0_bin = layers_path + "/bidirectional_3.bin";
std::string l1_bin = layers_path + "/bidirectional_4.bin";
std::string d0_bin = layers_path + "/dense_3.bin";
std::string d1_bin = layers_path + "/dense_4.bin";
net = new tk::dnn::Network(dim0);
tk::dnn::Input *x0 = new tk::dnn::Input (net, dim0, i0_d);
tk::dnn::Conv2d *x0_0 = new tk::dnn::Conv2d (net, 128, 1, 11, 1, 1, 0, 0, c0_bin);
tk::dnn::Conv2d *x0_1 = new tk::dnn::Conv2d (net, 128, 1, 11, 1, 1, 0, 0, c1_bin);
tk::dnn::Pooling *x0_2 = new tk::dnn::Pooling(net, 1, 3, 1, 3, tk::dnn::tkdnnPoolingMode_t::POOLING_MAX);
tk::dnn::Input *x1 = new tk::dnn::Input (net, dim1, i1_d);
tk::dnn::Conv2d *x1_0 = new tk::dnn::Conv2d (net, 128, 1, 11, 1, 1, 0, 0, c2_bin);
tk::dnn::Conv2d *x1_1 = new tk::dnn::Conv2d (net, 128, 1, 11, 1, 1, 0, 0, c3_bin);
tk::dnn::Pooling *x1_2 = new tk::dnn::Pooling(net, 1, 3, 1, 3, tk::dnn::tkdnnPoolingMode_t::POOLING_MAX);
tk::dnn::Input *x2 = new tk::dnn::Input (net, dim2, i2_d);
tk::dnn::Conv2d *x2_0 = new tk::dnn::Conv2d (net, 128, 1, 11, 1, 1, 0, 0, c4_bin);
tk::dnn::Conv2d *x2_1 = new tk::dnn::Conv2d (net, 128, 1, 11, 1, 1, 0, 0, c5_bin);
tk::dnn::Pooling *x2_2 = new tk::dnn::Pooling(net, 1, 3, 1, 3, tk::dnn::tkdnnPoolingMode_t::POOLING_MAX);
tk::dnn::Layer *concat_l[3] = { x0_2, x1_2, x2_2 };
tk::dnn::Route *concat = new tk::dnn::Route(net, concat_l, 3);
tk::dnn::LSTM *lstm0 = new tk::dnn::LSTM(net, 128, true, l0_bin);
tk::dnn::LSTM *lstm1 = new tk::dnn::LSTM(net, 128, false, l1_bin);
tk::dnn::Dense *d0 = new tk::dnn::Dense(net, 3, d0_bin);
tk::dnn::Layer *lstm1_l[1] = { lstm1 };
tk::dnn::Route *lstm1_link = new tk::dnn::Route(net, lstm1_l, 1);
tk::dnn::Dense *d1 = new tk::dnn::Dense(net, 4, d1_bin);
net->print();
// output data
o0_d = d0->dstData;
o1_d = d1->dstData;
odim0 = d0->output_dim;
odim1 = d1->output_dim;
deltaP.resize(1, odim0.tot());
deltaQ.resize(1, odim1.tot());
}
void update(dnnType *x0, dnnType *x1, dnnType *x2) {
checkCuda( cudaMemcpy(i0_d, x0, dim0.tot()*sizeof(dnnType), cudaMemcpyHostToDevice) );
checkCuda( cudaMemcpy(i1_d, x1, dim1.tot()*sizeof(dnnType), cudaMemcpyHostToDevice) );
checkCuda( cudaMemcpy(i2_d, x2, dim2.tot()*sizeof(dnnType), cudaMemcpyHostToDevice) );
// Inference
tk::dnn::dataDim_t dim;
net->infer(dim, nullptr);
checkCuda( cudaMemcpy(deltaP.data(), o0_d, odim0.tot()*sizeof(dnnType), cudaMemcpyDeviceToHost) );
checkCuda( cudaMemcpy(deltaQ.data(), o1_d, odim1.tot()*sizeof(dnnType), cudaMemcpyDeviceToHost) );
}
};
}}
include/tkDNN/Layer.h
+3 -1
View File
@@ -125,6 +125,7 @@ public:
virtual layerType_t getLayerType() { return LAYER_INPUT; };
virtual dnnType* infer(dataDim_t &dim, dnnType* srcData) {
dim = output_dim;
return dstData;
}
};
@@ -368,7 +369,8 @@ public:
virtual dnnType* infer(dataDim_t &dim, dnnType* srcData);
public:
Layer **layers; //ids of layers to be merged
static const int MAX_LAYERS = 32;
Layer *layers[MAX_LAYERS]; //ids of layers to be merged
int layers_n; //number of layers
};
include/tkDNN/utils.h
+1 -1
View File
@@ -88,7 +88,7 @@
} \
}
void printCenteredTitle(const char *title, char fill, int dim);
void printCenteredTitle(const char *title, char fill, int dim = 30);
bool fileExist(const char *fname);
void readBinaryFile(std::string fname, int size, dnnType** data_h, dnnType** data_d, int seek = 0);
int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device = true, int limit = 10);
src/Route.cpp
+8 -2
View File
@@ -7,9 +7,15 @@ namespace tk { namespace dnn {
Route::Route(Network *net, Layer **layers, int layers_n) : Layer(net) {
this->layers = layers;
// copy input layers
if(layers_n > MAX_LAYERS) {
FatalError("ROUTE: reached max number of input layers");
}
for(int i=0; i<layers_n; i++) {
this->layers[i] = layers[i];
}
this->layers_n = layers_n;
//get dims
output_dim.l = 1;
output_dim.c = 0;
src/utils.cpp
+8 -6
View File
@@ -53,8 +53,9 @@ void printDeviceVector(int size, dnnType* vec_d, bool device)
dnnType *vec;
if(device) {
vec = new dnnType[size];
cudaDeviceSynchronize();
cudaMemcpy(vec, vec_d, size*sizeof(dnnType), cudaMemcpyDeviceToHost);
checkCuda(cudaDeviceSynchronize());
checkCuda(cudaMemcpy(vec, vec_d, size*sizeof(dnnType), cudaMemcpyDeviceToHost));
checkCuda(cudaDeviceSynchronize());
} else {
vec = vec_d;
}
@@ -76,10 +77,11 @@ int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device, int
if(device) {
data_h = new dnnType[size];
correct_h = new dnnType[size];
cudaDeviceSynchronize();
cudaMemcpy(data_h, data_d, size*sizeof(dnnType), cudaMemcpyDeviceToHost);
cudaMemcpy(correct_h, correct_d, size*sizeof(dnnType), cudaMemcpyDeviceToHost);
checkCuda(cudaDeviceSynchronize());
checkCuda(cudaMemcpy(data_h, data_d, size*sizeof(dnnType), cudaMemcpyDeviceToHost));
checkCuda(cudaMemcpy(correct_h, correct_d, size*sizeof(dnnType), cudaMemcpyDeviceToHost));
checkCuda(cudaDeviceSynchronize());
} else {
data_h = data_d;
correct_h = correct_d;
tests/imuodom/imuodom.cpp
+35 -45
View File
@@ -1,5 +1,6 @@
#include<iostream>
#include "tkdnn.h"
#include "tkDNN/ImuOdom.h"
const char *i0_bin = "../tests/imuodom/layers/input0.bin";
const char *i1_bin = "../tests/imuodom/layers/input1.bin";
@@ -18,8 +19,14 @@ const char *l1_bin = "../tests/imuodom/layers/bidirectional_4.bin";
const char *d0_bin = "../tests/imuodom/layers/dense_3.bin";
const char *d1_bin = "../tests/imuodom/layers/dense_4.bin";
int main() {
tk::dnn::ImuOdom ImuNet;
ImuNet.init("../tests/imuodom/layers/");
const int N = 19513;
// Network layout
tk::dnn::dataDim_t dim0(1, 4, 1, 100);
tk::dnn::dataDim_t dim1(1, 3, 1, 100);
@@ -28,56 +35,39 @@ int main() {
// Load input
dnnType *i0_d, *i1_d, *i2_d;
dnnType *i0_h, *i1_h, *i2_h;
readBinaryFile(i0_bin, dim0.tot(), &i0_h, &i0_d);
readBinaryFile(i1_bin, dim1.tot(), &i1_h, &i1_d);
readBinaryFile(i2_bin, dim2.tot(), &i2_h, &i2_d);
tk::dnn::Network net(dim0);
tk::dnn::Input x0 (&net, dim0, i0_d);
tk::dnn::Conv2d x0_0(&net, 128, 1, 11, 1, 1, 0, 0, c0_bin);
tk::dnn::Conv2d x0_1(&net, 128, 1, 11, 1, 1, 0, 0, c1_bin);
tk::dnn::Pooling x0_2(&net, 1, 3, 1, 3, tk::dnn::tkdnnPoolingMode_t::POOLING_MAX);
tk::dnn::Input x1 (&net, dim1, i1_d);
tk::dnn::Conv2d x1_0(&net, 128, 1, 11, 1, 1, 0, 0, c2_bin);
tk::dnn::Conv2d x1_1(&net, 128, 1, 11, 1, 1, 0, 0, c3_bin);
tk::dnn::Pooling x1_2(&net, 1, 3, 1, 3, tk::dnn::tkdnnPoolingMode_t::POOLING_MAX);
tk::dnn::Input x2 (&net, dim2, i2_d);
tk::dnn::Conv2d x2_0(&net, 128, 1, 11, 1, 1, 0, 0, c4_bin);
tk::dnn::Conv2d x2_1(&net, 128, 1, 11, 1, 1, 0, 0, c5_bin);
tk::dnn::Pooling x2_2(&net, 1, 3, 1, 3, tk::dnn::tkdnnPoolingMode_t::POOLING_MAX);
tk::dnn::Layer *concat_l[3] = { &x0_2, &x1_2, &x2_2 };
tk::dnn::Route concat (&net, concat_l, 3);
tk::dnn::LSTM lstm0(&net, 128, true, l0_bin);
tk::dnn::LSTM lstm1(&net, 128, false, l1_bin);
tk::dnn::Dense d0 (&net, 3, d0_bin);
tk::dnn::Layer *lstm1_l[1] = { &lstm1 };
tk::dnn::Route lstm1_link (&net, lstm1_l, 1);
tk::dnn::Dense d1 (&net, 4, d1_bin);
net.print();
readBinaryFile(i0_bin, dim0.tot()*N, &i0_h, &i0_d);
readBinaryFile(i1_bin, dim1.tot()*N, &i1_h, &i1_d);
readBinaryFile(i2_bin, dim2.tot()*N, &i2_h, &i2_d);
dnnType *data;
tk::dnn::dataDim_t dim;
TIMER_START
// Inference
data = net.infer(dim, data);
TIMER_STOP
// Print real test
std::cout<<"\n==== CHECK RESULT ====\n";
dnnType *out0, *out1;
dnnType *out0_h, *out1_h;
readBinaryFile(o0_bin, d0.output_dim.tot(), &out0_h, &out0);
readBinaryFile(o1_bin, d1.output_dim.tot(), &out1_h, &out1);
d0.output_dim.print();
checkResult(d0.output_dim.tot(), d0.dstData, out0);
d1.output_dim.print();
checkResult(d1.output_dim.tot(), d1.dstData, out1);
readBinaryFile(o0_bin, ImuNet.odim0.tot()*N, &out0_h, &out0);
readBinaryFile(o1_bin, ImuNet.odim1.tot()*N, &out1_h, &out1);
for(int i=0; i<N; i++) {
TIMER_START
// Inference
ImuNet.update(i0_h, i1_h, i2_h);
TIMER_STOP
// Print real test
printCenteredTitle( (std::string(" CHECK RESULT ") + std::to_string(i) + " ").c_str() , '=');
//ImuNet.odim0.print();
checkResult(ImuNet.odim0.tot(), out0, ImuNet.o0_d);
//ImuNet.odim1.print();
checkResult(ImuNet.odim0.tot(), out1, ImuNet.o1_d);
i0_h += ImuNet.dim0.tot();
i1_h += ImuNet.dim1.tot();
i2_h += ImuNet.dim2.tot();
out0 += ImuNet.odim0.tot();
out1 += ImuNet.odim1.tot();
}
return 0;
}
tests/imuodom/infer.py
+18 -9
View File
@@ -8,6 +8,7 @@ import os
import random
import struct
from keras.models import Sequential, Model
import pickle
def bin_write(f, data):
data = data.flatten()
@@ -24,15 +25,23 @@ if __name__ == '__main__':
print("Load model: ", "ferrariS1.hdf5")
model = load_model("ferrariS1.hdf5")
model.summary()
weights = model.get_weights()
np.random.seed(2)
x_angle = np.random.rand(1,100,4)
x_gyro = np.random.rand(1,100,3)
x_acc = np.random.rand(1,100,3)
indata = pickle.load(open("input.pk", 'rb'))
outdata = pickle.load(open("output.pk", 'rb'))
x_angle = indata[0]
x_gyro = indata[1]
x_acc = indata[2]
[yhat_delta_p, yhat_delta_q] = model.predict(indata, batch_size=1, verbose=1)
predictdata = [yhat_delta_p, yhat_delta_q]
error = outdata[0] - predictdata[0]
print("error delta_p: ", error.sum())
error = outdata[1] - predictdata[1]
print("error delta_q: ", error.sum())
[yhat_delta_p, yhat_delta_q] = model.predict([x_angle, x_gyro, x_acc], batch_size=1, verbose=1)
#layer_name = 'dense_4'
#intermediate_layer_model = Model(inputs=model.input,
@@ -45,9 +54,9 @@ if __name__ == '__main__':
x_acc = np.array([x_acc])
#intermediate_output = np.array([intermediate_output])
x_angle = x_angle.transpose(0, 3, 1, 2)
x_gyro = x_gyro.transpose(0, 3, 1, 2)
x_acc = x_acc.transpose(0, 3, 1, 2)
x_angle = x_angle.transpose(1, 3, 0, 2)
x_gyro = x_gyro.transpose(1, 3, 0, 2)
x_acc = x_acc.transpose(1, 3, 0, 2)
#intermediate_output = intermediate_output.transpose(0, 3, 1, 2)
#print("Aggregate:")
#print(intermediate_output.tolist())