augustin64/tipe
1
1
Fork 0
mirror of https://github.com/augustin64/projet-tipe synced 2025-01-24 07:36:24 +01:00
Code Issues Packages Projects Releases Wiki Activity

Changes in the structure of the cnn

Browse Source
This commit is contained in:
Julien Chemillier 2022-09-19 18:39:49 +02:00
parent d236055a6d
commit cbd7b5427d
5 changed files with 106 additions and 41 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
src/cnn/creation.c
View File

@ -14,7 +14,7 @@ Network* create_network(int max_size, int dropout, int initialisation, int input
network->initialisation = initialisation; network->initialisation = initialisation;
network->size = 1; network->size = 1;
network->input = (float****)malloc(sizeof(float***)*max_size); network->input = (float****)malloc(sizeof(float***)*max_size);
network->kernel = (Kernel**)malloc(sizeof(Kernel*)*(max_size-1)); network->kernel = (Kernel**)malloc(sizeof(Kernel*)*max_size);
network->width = (int*)malloc(sizeof(int*)*max_size); network->width = (int*)malloc(sizeof(int*)*max_size);
network->depth = (int*)malloc(sizeof(int*)*max_size); network->depth = (int*)malloc(sizeof(int*)*max_size);
for (int i=0; i < max_size; i++) { for (int i=0; i < max_size; i++) {
@ -32,12 +32,12 @@ Network* create_network_lenet5(int dropout, int activation, int initialisation)
Network* network = create_network(8, dropout, initialisation, 32, 1); Network* network = create_network(8, dropout, initialisation, 32, 1);
network->kernel[0]->activation = activation; network->kernel[0]->activation = activation;
add_convolution(network, 6, 5, activation); add_convolution(network, 6, 5, activation);
add_average_pooling(network, 2, activation); add_2d_average_pooling(network, 2);
add_convolution(network, 16, 5, activation); add_convolution(network, 16, 5, activation);
add_average_pooling_flatten(network, 2, activation); add_2d_average_pooling(network, 2);
add_dense_linearisation(network, 160, 120, activation);
add_dense(network, 120, 84, activation); add_dense(network, 120, 84, activation);
add_dense(network, 84, 10, activation); add_dense(network, 84, 10, SOFTMAX);
add_dense(network, 10, 10, SOFTMAX);
return network; return network;
} }
@ -57,9 +57,11 @@ void create_a_line_input_layer(Network* network, int pos, int dim) {
network->input[pos] = (float***)malloc(sizeof(float**)); network->input[pos] = (float***)malloc(sizeof(float**));
network->input[pos][0] = (float**)malloc(sizeof(float*)); network->input[pos][0] = (float**)malloc(sizeof(float*));
network->input[pos][0][0] = (float*)malloc(sizeof(float)*dim); network->input[pos][0][0] = (float*)malloc(sizeof(float)*dim);
network->width[pos] = dim;
network->depth[pos] = 1;
} }
void add_average_pooling(Network* network, int kernel_size, int activation) { void add_2d_average_pooling(Network* network, int kernel_size) {
int n = network->size; int n = network->size;
if (network->max_size == n) { if (network->max_size == n) {
printf("Impossible de rajouter une couche d'average pooling, le réseau est déjà plein\n"); printf("Impossible de rajouter une couche d'average pooling, le réseau est déjà plein\n");
@ -67,12 +69,12 @@ void add_average_pooling(Network* network, int kernel_size, int activation) {
} }
network->kernel[n]->cnn = NULL; network->kernel[n]->cnn = NULL;
network->kernel[n]->nn = NULL; network->kernel[n]->nn = NULL;
network->kernel[n]->activation = activation + 100*kernel_size; network->kernel[n]->activation = 100*kernel_size; // Ne contient pas de fonction d'activation
create_a_cube_input_layer(network, n, network->depth[n-1], network->width[n-1]/2); create_a_cube_input_layer(network, n, network->depth[n-1], network->width[n-1]/2);
network->size++; network->size++;
} }
void add_average_pooling_flatten(Network* network, int kernel_size, int activation) { void add_average_pooling_flatten(Network* network, int kernel_size) { // NEED TO BE VERIFIED
int n = network->size; int n = network->size;
if (network->max_size == n) { if (network->max_size == n) {
printf("Impossible de rajouter une couche d'average pooling, le réseau est déjà plein\n"); printf("Impossible de rajouter une couche d'average pooling, le réseau est déjà plein\n");
@ -80,7 +82,7 @@ void add_average_pooling_flatten(Network* network, int kernel_size, int activati
} }
network->kernel[n]->cnn = NULL; network->kernel[n]->cnn = NULL;
network->kernel[n]->nn = NULL; network->kernel[n]->nn = NULL;
network->kernel[n]->activation = activation + 100*kernel_size; network->kernel[n]->activation = 100*kernel_size; // Ne contient pas de fonction d'activation
int dim = (network->width[n-1]*network->width[n-1]*network->depth[n-1])/(kernel_size*kernel_size); int dim = (network->width[n-1]*network->width[n-1]*network->depth[n-1])/(kernel_size*kernel_size);
create_a_line_input_layer(network, n, dim); create_a_line_input_layer(network, n, dim);
network->size++; network->size++;
@ -92,6 +94,7 @@ void add_convolution(Network* network, int depth_output, int kernel_size, int ac
printf("Impossible de rajouter une couche de convolution, le réseau est déjà plein \n"); printf("Impossible de rajouter une couche de convolution, le réseau est déjà plein \n");
return; return;
} }
int bias_size = network->width[n-1] - 2*(kernel_size/2);
int depth_input = network->depth[n-1]; int depth_input = network->depth[n-1];
network->kernel[n]->nn = NULL; network->kernel[n]->nn = NULL;
network->kernel[n]->activation = activation; network->kernel[n]->activation = activation;
@ -118,14 +121,14 @@ void add_convolution(Network* network, int depth_output, int kernel_size, int ac
cnn->bias = (float***)malloc(sizeof(float**)*depth_output); cnn->bias = (float***)malloc(sizeof(float**)*depth_output);
cnn->d_bias = (float***)malloc(sizeof(float**)*depth_output); cnn->d_bias = (float***)malloc(sizeof(float**)*depth_output);
for (int i=0; i < depth_output; i++) { for (int i=0; i < depth_output; i++) {
cnn->bias[i] = (float**)malloc(sizeof(float*)*kernel_size); cnn->bias[i] = (float**)malloc(sizeof(float*)*bias_size);
cnn->d_bias[i] = (float**)malloc(sizeof(float*)*kernel_size); cnn->d_bias[i] = (float**)malloc(sizeof(float*)*bias_size);
for (int j=0; j < kernel_size; j++) { for (int j=0; j < bias_size; j++) {
cnn->bias[i][j] = (float*)malloc(sizeof(float)*kernel_size); cnn->bias[i][j] = (float*)malloc(sizeof(float)*bias_size);
cnn->d_bias[i][j] = (float*)malloc(sizeof(float)*kernel_size); cnn->d_bias[i][j] = (float*)malloc(sizeof(float)*bias_size);
} }
} }
create_a_cube_input_layer(network, n, depth_output, network->width[n-1] - 2*(kernel_size/2)); create_a_cube_input_layer(network, n, depth_output, bias_size);
int n_int = network->width[n-1]*network->width[n-1]*network->depth[n-1]; int n_int = network->width[n-1]*network->width[n-1]*network->depth[n-1];
int n_out = network->width[n]*network->width[n]*network->depth[n]; int n_out = network->width[n]*network->width[n]*network->depth[n];
initialisation_3d_matrix(network->initialisation, cnn->bias, depth_output, kernel_size, kernel_size, n_int+n_out); initialisation_3d_matrix(network->initialisation, cnn->bias, depth_output, kernel_size, kernel_size, n_int+n_out);
@ -162,3 +165,35 @@ void add_dense(Network* network, int input_units, int output_units, int activati
create_a_line_input_layer(network, n, output_units); create_a_line_input_layer(network, n, output_units);
network->size++; network->size++;
} }
void add_dense_linearisation(Network* network, int input_units, int output_units, int activation) {
// Can replace input_units by a research of this dim
int n = network->size;
if (network->max_size == n) {
printf("Impossible de rajouter une couche dense, le réseau est déjà plein\n");
return;
}
network->kernel[n]->cnn = NULL;
network->kernel[n]->nn = (Kernel_nn*)malloc(sizeof(Kernel_nn));
Kernel_nn* nn = network->kernel[n]->nn;
network->kernel[n]->activation = activation;
nn->input_units = input_units;
nn->output_units = output_units;
nn->bias = (float*)malloc(sizeof(float)*output_units);
nn->d_bias = (float*)malloc(sizeof(float)*output_units);
nn->weights = (float**)malloc(sizeof(float*)*input_units);
nn->d_weights = (float**)malloc(sizeof(float*)*input_units);
for (int i=0; i < input_units; i++) {
nn->weights[i] = (float*)malloc(sizeof(float)*output_units);
nn->d_weights[i] = (float*)malloc(sizeof(float)*output_units);
}
initialisation_1d_matrix(network->initialisation, nn->bias, output_units, output_units+input_units);
initialisation_1d_matrix(ZERO, nn->d_bias, output_units, output_units+input_units);
initialisation_2d_matrix(network->initialisation, nn->weights, input_units, output_units, output_units+input_units);
initialisation_2d_matrix(ZERO, nn->d_weights, input_units, output_units, output_units+input_units);
create_a_line_input_layer(network, n, output_units);
network->size++;
}

13
src/cnn/include/creation.h
View File

@ -26,21 +26,26 @@ void create_a_line_input_layer(Network* network, int pos, int dim);
/* /*
* Ajoute au réseau une couche d'average pooling valide de dimension dim*dim * Ajoute au réseau une couche d'average pooling valide de dimension dim*dim
*/ */
void add_average_pooling(Network* network, int kernel_size, int activation); void add_2d_average_pooling(Network* network, int kernel_size);
/* /*
* Ajoute au réseau une couche d'average pooling valide de dimension dim*dim qui aplatit * Ajoute au réseau une couche d'average pooling valide de dimension dim*dim qui aplatit
*/ */
void add_average_pooling_flatten(Network* network, int kernel_size, int activation); void add_average_pooling_flatten(Network* network, int kernel_size);
/* /*
* Ajoute une couche de convolution dim*dim au réseau et initialise les kernels * Ajoute au réseau une couche de convolution dim*dim et initialise les kernels
*/ */
void add_convolution(Network* network, int nb_filter, int kernel_size, int activation); void add_convolution(Network* network, int nb_filter, int kernel_size, int activation);
/* /*
* Ajoute une couche dense au réseau et initialise les poids et les biais * Ajoute au réseau une couche dense et initialise les poids et les biais
*/ */
void add_dense(Network* network, int input_units, int output_units, int activation); void add_dense(Network* network, int input_units, int output_units, int activation);
/*
* Ajoute au réseau une couche dense qui aplatit
*/
void add_dense_linearisation(Network* network, int input_units, int output_units, int activation);
#endif #endif

3
src/cnn/include/struct.h
View File

@ -23,7 +23,8 @@ typedef struct Kernel_nn {
typedef struct Kernel { typedef struct Kernel {
Kernel_cnn* cnn; // NULL si ce n'est pas un cnn Kernel_cnn* cnn; // NULL si ce n'est pas un cnn
Kernel_nn* nn; // NULL si ce n'est pas un nn Kernel_nn* nn; // NULL si ce n'est pas un nn
int activation; // Vaut l'activation sauf pour un pooling où il: vaut kernel_size*100 + activation int activation; // Vaut l'activation sauf pour un pooling où il: vaut pooling_size*100 + activation
int linearisation; // Vaut 1 si c'est la linéarisation d'une couche, 0 sinon ?? Ajouter dans les autres
} Kernel; } Kernel;

62
src/cnn/main.c
View File

@ -30,35 +30,48 @@ void write_image_in_network_32(int** image, int height, int width, float** input
} }
void forward_propagation(Network* network) { void forward_propagation(Network* network) {
int output_dim, output_depth; int activation, input_width, input_depth, output_width, output_depth;
int n = network->size;
float*** input;
float*** output; float*** output;
for (int i=0; i < network->size-1; i++) { Kernel* k_i_1;
if (network->kernel[i]->nn==NULL && network->kernel[i]->cnn!=NULL) { //CNN Kernel* k_i;
output = network->input[i+1]; for (int i=0; i < n-1; i++) {
output_dim = network->width[i+1]; k_i_1 = network->kernel[i+1];
output_depth = network->depth[i+1]; k_i = network->kernel[i];
make_convolution(network->input[i], network->kernel[i]->cnn, output, output_dim); input_width = network->width[i];
choose_apply_function_input(network->kernel[i]->activation, output, output_depth, output_dim, output_dim); input_depth = network->depth[i];
output_width = network->width[i+1];
output_depth = network->depth[i+1];
activation = network->kernel[i]->activation;
input = network->input[i];
output = network->input[i+1];
if (k_i_1->nn==NULL && k_i_1->cnn!=NULL) { //CNN
printf("Convolution of cnn: %dx%d -> %dx%d\n", input_depth, input_width, output_depth, output_width);
make_convolution(input, k_i_1->cnn, output, output_width);
choose_apply_function_input(activation, output, output_depth, output_width, output_width);
} }
else if (network->kernel[i]->nn!=NULL && network->kernel[i]->cnn==NULL) { //NN else if (k_i_1->nn!=NULL && k_i_1->cnn==NULL) { //NN
make_fully_connected(network->input[i][0][0], network->kernel[i]->nn, network->input[i+1][0][0], network->width[i], network->width[i+1]); printf("Densification of nn\n");
choose_apply_function_input(network->kernel[i]->activation, network->input[i+1], 1, 1, network->width[i+1]); // Checked if it is a nn which linearise
make_fully_connected(network->input[i][0][0], network->kernel[i]->nn, network->input[i+1][0][0], input_width, output_width);
choose_apply_function_input(activation, output, 1, 1, output_width);
} }
else { //Pooling else { //Pooling (Vérifier dedans) ??
if (network->size-2==i) { if (n-2==i) {
printf("Le réseau ne peut pas finir par une pooling layer"); printf("Le réseau ne peut pas finir par une pooling layer");
return; return;
} }
if (network->kernel[i+1]->nn!=NULL && network->kernel[i+1]->cnn==NULL) { if (1==1) { // Pooling sur une matrice
make_average_pooling_flattened(network->input[i], network->input[i+1][0][0], network->kernel[i]->activation/100, network->depth[i], network->width[i]); printf("Average pooling\n");
choose_apply_function_input(network->kernel[i]->activation%100, network->input[i+1], 1, 1, network->width[i+1]); make_average_pooling(input, output, activation/100, output_depth, output_width);
} }
else if (network->kernel[i+1]->nn==NULL && network->kernel[i+1]->cnn!=NULL) { else if (1==0) { // Pooling sur un vecteur
make_average_pooling(network->input[i], network->input[i+1], network->kernel[i]->activation/100, network->depth[i+1], network->width[i+1]); printf("Error: Not implemented: forward: %d\n", i);
choose_apply_function_input(network->kernel[i]->activation%100, network->input[i+1], network->depth[i+1], network->width[i+1], network->width[i+1]);
} }
else { else {
printf("Le réseau ne peut pas contenir deux pooling layers collées"); printf("Erreur: forward_propagation: %d -> %d %d\n", i, k_i_1->nn==NULL, k_i_1->cnn);
return; return;
} }
} }
@ -128,6 +141,15 @@ float* generate_wanted_output(float wanted_number) {
int main() { int main() {
Network* network = create_network_lenet5(0, TANH, GLOROT_NORMAL); Network* network = create_network_lenet5(0, TANH, GLOROT_NORMAL);
for (int i=0; i<8; i++) {
printf("%d %d \n", network->depth[i], network->width[i]);
}
printf("Kernel:\n");
for (int i=0; i<7; i++) {
if (network->kernel[i]->cnn!=NULL) {
printf("%d -> %d %d\n", i, network->kernel[i]->cnn->rows, network->kernel[i]->cnn->k_size);
}
}
forward_propagation(network); forward_propagation(network);
return 0; return 0;
} }

4
src/cnn/make.c
View File

@ -5,6 +5,8 @@ void make_convolution(float*** input, Kernel_cnn* kernel, float*** output, int o
//NOT FINISHED, MISS CONDITIONS ON THE CONVOLUTION //NOT FINISHED, MISS CONDITIONS ON THE CONVOLUTION
float f; float f;
int n = kernel->k_size; int n = kernel->k_size;
printf("Convolution output: %dx%dx%d, %dx%dx%d\n", kernel->columns, output_dim, output_dim, kernel->rows, n, n);
printf("BIS %d %d \n", kernel->columns, kernel->k_size);
for (int i=0; i < kernel->columns; i++) { for (int i=0; i < kernel->columns; i++) {
for (int j=0; j < output_dim; j++) { for (int j=0; j < output_dim; j++) {
for (int k=0; k < output_dim; k++) { for (int k=0; k < output_dim; k++) {
@ -35,7 +37,7 @@ void make_average_pooling(float*** input, float*** output, int size, int output_
average += input[i][2*j +a][2*k +b]; average += input[i][2*j +a][2*k +b];
} }
} }
output[i][j][k] = average; output[i][j][k] = average/n;
} }
} }
} }