augustin64/tipe
1
1
Fork 0
mirror of https://github.com/augustin64/projet-tipe synced 2025-01-23 23:26:25 +01:00
Code Issues Packages Projects Releases Wiki Activity

Compare commits

base: augustin64:4625ad29923dc7010eb8dac93c74684615f2b9f8
Branches Tags
augustin64:main
..
compare: augustin64:0fb23c9b159cabb041a949d9884bbcd7aebd2caf
Branches Tags
augustin64:main

No commits in common. "4625ad29923dc7010eb8dac93c74684615f2b9f8" and "0fb23c9b159cabb041a949d9884bbcd7aebd2caf" have entirely different histories.
4625ad2992 ... 0fb23c9b15

13 changed files with 127 additions and 183 deletions
Show Stats Expand all files Collapse all files

82
Makefile
View File

@ -41,12 +41,7 @@ NVCCFLAGS = -g
# -fsanitize=address -lasan
#! WARNING: test/cnn-neuron_io fails with this option enabled
all: dense cnn;
#
# Build dense
#
@ -66,95 +61,82 @@ $(BUILDDIR)/dense_%.o: $(DENSE_SRCDIR)/%.c $(DENSE_SRCDIR)/include/%.h
$(CC) -c $< -o $@ $(CFLAGS)
#
# Build cnn
#
cnn: $(BUILDDIR)/cnn-main $(BUILDDIR)/cnn-main-cuda $(BUILDDIR)/cnn-preview $(BUILDDIR)/cnn-export;
$(BUILDDIR)/cnn-main: $(CNN_SRCDIR)/main.c \
$(BUILDDIR)/cnn_backpropagation.o \
$(BUILDDIR)/cnn_initialisation.o \
$(BUILDDIR)/cnn_train.o \
$(BUILDDIR)/cnn_test_network.o \
$(BUILDDIR)/cnn_convolution.o \
$(BUILDDIR)/cnn_cnn.o \
$(BUILDDIR)/cnn_creation.o \
$(BUILDDIR)/cnn_initialisation.o \
$(BUILDDIR)/cnn_make.o \
$(BUILDDIR)/cnn_neuron_io.o \
$(BUILDDIR)/cnn_function.o \
$(BUILDDIR)/cnn_creation.o \
$(BUILDDIR)/cnn_models.o \
$(BUILDDIR)/cnn_update.o \
$(BUILDDIR)/cnn_train.o \
$(BUILDDIR)/cnn_utils.o \
$(BUILDDIR)/cnn_make.o \
$(BUILDDIR)/cnn_update.o \
$(BUILDDIR)/cnn_free.o \
$(BUILDDIR)/cnn_jpeg.o \
$(BUILDDIR)/cnn_cnn.o \
\
$(BUILDDIR)/cnn_convolution.o \
$(BUILDDIR)/cnn_backpropagation.o \
$(BUILDDIR)/memory_management.o \
$(BUILDDIR)/colors.o \
$(BUILDDIR)/mnist.o \
$(BUILDDIR)/utils.o
$(CC) $^ -o $@ $(CFLAGS) $(LD_CFLAGS)
ifdef NVCC_INSTALLED
$(BUILDDIR)/cnn-main-cuda: $(BUILDDIR)/cnn_main.cuda.o \
$(BUILDDIR)/cnn_initialisation.cuda.o \
$(BUILDDIR)/cnn_test_network.cuda.o \
$(BUILDDIR)/cnn_neuron_io.cuda.o \
$(BUILDDIR)/cnn_creation.cuda.o \
$(BUILDDIR)/cnn_models.cuda.o \
$(BUILDDIR)/cnn_update.cuda.o \
$(BUILDDIR)/cnn_train.cuda.o \
$(BUILDDIR)/cnn_test_network.cuda.o \
$(BUILDDIR)/cnn_cnn.cuda.o \
$(BUILDDIR)/cnn_creation.cuda.o \
$(BUILDDIR)/cnn_initialisation.cuda.o \
$(BUILDDIR)/cnn_cuda_make.o \
$(BUILDDIR)/cnn_neuron_io.cuda.o \
$(BUILDDIR)/cnn_cuda_function.o \
$(BUILDDIR)/cnn_utils.cuda.o \
$(BUILDDIR)/cnn_update.cuda.o \
$(BUILDDIR)/cnn_free.cuda.o \
$(BUILDDIR)/cnn_jpeg.cuda.o \
$(BUILDDIR)/cnn_cnn.cuda.o \
\
$(BUILDDIR)/cnn_cuda_backpropagation.o \
$(BUILDDIR)/cnn_cuda_convolution.o \
$(BUILDDIR)/cnn_cuda_function.o \
$(BUILDDIR)/cnn_cuda_make.o \
\
$(BUILDDIR)/cuda_memory_management.o \
$(BUILDDIR)/cnn_cuda_backpropagation.o \
$(BUILDDIR)/colors.cuda.o \
$(BUILDDIR)/cuda_memory_management.o \
$(BUILDDIR)/mnist.cuda.o \
$(BUILDDIR)/cuda_utils.o
$(BUILDDIR)/cuda_utils.o
$(NVCC) $(LD_NVCCFLAGS) $(NVCCFLAGS) $^ -o $@
else
$(BUILDDIR)/cnn-main-cuda:
@echo "$(NVCC) not found, skipping"
endif
$(BUILDDIR)/cnn-preview: $(CNN_SRCDIR)/preview.c $(BUILDDIR)/cnn_jpeg.o $(BUILDDIR)/colors.o $(BUILDDIR)/utils.o
$(CC) $^ -o $@ $(CFLAGS) $(LD_CFLAGS)
$(BUILDDIR)/cnn-export: $(CNN_SRCDIR)/export.c \
$(BUILDDIR)/cnn_backpropagation.o \
$(BUILDDIR)/cnn_convolution.o \
$(BUILDDIR)/cnn_neuron_io.o \
$(BUILDDIR)/cnn_function.o \
$(BUILDDIR)/cnn_free.o \
$(BUILDDIR)/cnn_make.o \
$(BUILDDIR)/cnn_cnn.o \
$(BUILDDIR)/cnn_jpeg.o \
\
$(BUILDDIR)/cnn_neuron_io.o \
$(BUILDDIR)/utils.o \
$(BUILDDIR)/memory_management.o \
$(BUILDDIR)/cnn_cnn.o \
$(BUILDDIR)/cnn_make.o \
$(BUILDDIR)/cnn_backpropagation.o \
$(BUILDDIR)/cnn_function.o \
$(BUILDDIR)/cnn_convolution.o \
$(BUILDDIR)/colors.o \
$(BUILDDIR)/mnist.o \
$(BUILDDIR)/utils.o
$(BUILDDIR)/cnn_jpeg.o
$(CC) $^ -o $@ $(CFLAGS) $(LD_CFLAGS)
$(BUILDDIR)/cnn_%.o: $(CNN_SRCDIR)/%.c $(CNN_SRCDIR)/include/%.h
$(CC) -c $< -o $@ $(CFLAGS)
$(BUILDDIR)/cnn_%.cuda.o: $(CNN_SRCDIR)/%.c $(CNN_SRCDIR)/include/%.h
$(CC) -c $< -o $@ $(CFLAGS) -DUSE_CUDA -lcuda -I$(CUDA_INCLUDE)
ifdef NVCC_INSTALLED
$(BUILDDIR)/cnn_cuda_%.o: $(CNN_SRCDIR)/%.cu $(CNN_SRCDIR)/include/%.h
$(NVCC) $(NVCCFLAGS) -c -dc $< -o $@
@ -162,20 +144,15 @@ else
$(BUILDDIR)/cnn_cuda_%.o: $(CNN_SRCDIR)/%.cu $(CNN_SRCDIR)/include/%.h
@echo "$(NVCC) not found, skipping"
endif
#
# Build general files
#
$(BUILDDIR)/%.o: $(COMMON_SRCDIR)/%.c $(COMMON_SRCDIR)/include/%.h
$(CC) -c $< -o $@ $(CFLAGS)
$(BUILDDIR)/%.cuda.o: $(COMMON_SRCDIR)/%.c $(COMMON_SRCDIR)/include/%.h
$(CC) -c $< -o $@ $(CFLAGS) -DUSE_CUDA -lcuda -I$(CUDA_INCLUDE)
ifdef NVCC_INSTALLED
$(BUILDDIR)/cuda_%.o: $(COMMON_SRCDIR)/%.cu $(COMMON_SRCDIR)/include/%.h
$(NVCC) $(NVCCFLAGS) -c -dc $< -o $@
@ -183,8 +160,6 @@ else
@echo "$(NVCC) not found, skipping"
endif
#
# Tests
#
@ -232,8 +207,6 @@ $(BUILDDIR)/test-cnn_%: $(TEST_SRCDIR)/cnn_%.cu
@echo "$(NVCC) not found, skipping"
endif
#
# Utils
#
@ -258,7 +231,6 @@ $(CACHE_DIR)/mnist-reseau-cnn.bin: $(BUILDDIR)/cnn-main
--out $(CACHE_DIR)/mnist-reseau-cnn.bin
#
# Clean project
#

4
doc/cnn/neuron_io.md
View File

@ -38,8 +38,6 @@ uint32_t|linearisation|
uint32_t|k_size|
uint32_t|rows|
uint32_t|columns|
uint32_t|stride|
uint32_t|padding|
#### Si la couche est un nn:
type | nom de la variable | commentaire
@ -54,8 +52,6 @@ type | nom de la variable | commentaire
:---:|:---:|:---:
uint32_t|linearisation|
uint32_t|pooling|
uint32_t|stride|
uint32_t|padding|
### Corps

67
src/cnn/creation.c
View File

@ -34,6 +34,73 @@ Network* create_network(int max_size, float learning_rate, int dropout, int init
return network;
}
Network* create_network_lenet5(float learning_rate, int dropout, int activation, int initialisation, int input_width, int input_depth) {
Network* network = create_network(8, learning_rate, dropout, initialisation, input_width, input_depth);
add_convolution(network, 5, 6, 1, 0, activation);
add_average_pooling(network, 2, 2, 0);
add_convolution(network, 5, 16, 1, 0, activation);
add_average_pooling(network, 2, 2, 0);
add_dense_linearisation(network, 120, activation);
add_dense(network, 84, activation);
add_dense(network, 10, SOFTMAX);
return network;
}
Network* create_network_alexnet(float learning_rate, int dropout, int activation, int initialisation, int size_output) {
Network* network = create_network(12, learning_rate, dropout, initialisation, 227, 3);
add_convolution(network, 11, 96, 4, 0, activation);
add_average_pooling(network, 3, 2, 0);
add_convolution(network, 5, 256, 1, 2, activation);
add_average_pooling(network, 3, 2, 0);
add_convolution(network, 3, 384, 1, 1, activation);
add_convolution(network, 3, 384, 1, 1, activation);
add_convolution(network, 3, 256, 1, 1, activation);
add_average_pooling(network, 3, 2, 0);
add_dense_linearisation(network, 4096, activation);
add_dense(network, 4096, activation);
add_dense(network, size_output, SOFTMAX);
return network;
}
Network* create_network_VGG16(float learning_rate, int dropout, int activation, int initialisation, int size_output) {
Network* network = create_network(23, learning_rate, dropout, initialisation, 256, 3);
add_convolution(network, 3, 64, 1, 0, activation); // Conv3-64
add_convolution(network, 3, 64, 1, 0, activation); // Conv3-64
add_average_pooling(network, 2, 2, 0); // Max Pool
add_convolution(network, 3, 128, 1, 0, activation); // Conv3-128
add_convolution(network, 1, 128, 1, 0, activation); // Conv1-128
add_average_pooling(network, 2, 2, 0); // Max Pool
add_convolution(network, 3, 256, 1, 0, activation); // Conv3-256
add_convolution(network, 3, 256, 1, 0, activation); // Conv3-256
add_convolution(network, 1, 256, 1, 0, activation); // Conv1-256
add_average_pooling(network, 2, 2, 0); // Max Pool
add_convolution(network, 3, 512, 1, 0, activation); // Conv3-512
add_convolution(network, 3, 512, 1, 0, activation); // Conv3-512
add_convolution(network, 1, 512, 1, 0, activation); // Conv1-512
add_average_pooling(network, 2, 2, 0); // Max Pool
add_convolution(network, 3, 512, 1, 0, activation); // Conv3-512
add_convolution(network, 3, 512, 1, 0, activation); // Conv3-512
add_convolution(network, 1, 512, 1, 0, activation); // Conv1-512
add_average_pooling(network, 2, 2, 0); // Max Pool
add_dense_linearisation(network, 2048, activation);
add_dense(network, 2048, activation);
add_dense(network, 256, activation);
add_dense(network, size_output, SOFTMAX);
return network;
}
Network* create_simple_one(float learning_rate, int dropout, int activation, int initialisation, int input_width, int input_depth) {
Network* network = create_network(3, learning_rate, dropout, initialisation, input_width, input_depth);
add_dense_linearisation(network, 80, activation);
add_dense(network, 10, SOFTMAX);
return network;
}
void create_a_cube_input_layer(Network* network, int pos, int depth, int dim) {
network->input[pos] = (float***)nalloc(depth, sizeof(float**));
for (int i=0; i < depth; i++) {

22
src/cnn/include/creation.h
View File

@ -9,6 +9,28 @@
*/
Network* create_network(int max_size, float learning_rate, int dropout, int initialisation, int input_width, int input_depth);
/*
* Renvoie un réseau suivant l'architecture LeNet5
*/
Network* create_network_lenet5(float learning_rate, int dropout, int activation, int initialisation, int input_width, int input_depth);
/*
* Renvoie un réseau suivant l'architecture AlexNet
* C'est à dire en entrée 3x227x227 et une sortie de taille 'size_output'
*/
Network* create_network_alexnet(float learning_rate, int dropout, int activation, int initialisation, int size_output);
/*
* Renvoie un réseau suivant l'architecture VGG16 modifiée pour prendre en entrée 3x256x256
* et une sortie de taille 'size_output'
*/
Network* create_network_VGG16(float learning_rate, int dropout, int activation, int initialisation, int size_output);
/*
* Renvoie un réseau sans convolution, similaire à celui utilisé dans src/dense
*/
Network* create_simple_one(float learning_rate, int dropout, int activation, int initialisation, int input_width, int input_depth);
/*
* Créé et alloue de la mémoire à une couche de type input cube
*/

29
src/cnn/include/models.h
View File

@ -1,29 +0,0 @@
#include <stdlib.h>
#include <stdio.h>
#include "struct.h"
#ifndef DEF_MODELS_H
#define DEF_MODELS_H
/*
* Renvoie un réseau suivant l'architecture LeNet5
*/
Network* create_network_lenet5(float learning_rate, int dropout, int activation, int initialisation, int input_width, int input_depth);
/*
* Renvoie un réseau suivant l'architecture AlexNet
* C'est à dire en entrée 3x227x227 et une sortie de taille 'size_output'
*/
Network* create_network_alexnet(float learning_rate, int dropout, int activation, int initialisation, int size_output);
/*
* Renvoie un réseau suivant l'architecture VGG16 modifiée pour prendre en entrée 3x256x256
* et une sortie de taille 'size_output'
*/
Network* create_network_VGG16(float learning_rate, int dropout, int activation, int initialisation, int size_output);
/*
* Renvoie un réseau sans convolution, similaire à celui utilisé dans src/dense
*/
Network* create_simple_one(float learning_rate, int dropout, int activation, int initialisation, int input_width, int input_depth);
#endif

75
src/cnn/models.c
View File

@ -1,75 +0,0 @@
#include <stdlib.h>
#include <stdio.h>
#include "include/creation.h"
#include "include/function.h"
#include "include/struct.h"
#include "include/models.h"
Network* create_network_lenet5(float learning_rate, int dropout, int activation, int initialisation, int input_width, int input_depth) {
Network* network = create_network(8, learning_rate, dropout, initialisation, input_width, input_depth);
add_convolution(network, 5, 6, 1, 0, activation);
add_average_pooling(network, 2, 2, 0);
add_convolution(network, 5, 16, 1, 0, activation);
add_average_pooling(network, 2, 2, 0);
add_dense_linearisation(network, 120, activation);
add_dense(network, 84, activation);
add_dense(network, 10, SOFTMAX);
return network;
}
Network* create_network_alexnet(float learning_rate, int dropout, int activation, int initialisation, int size_output) {
Network* network = create_network(12, learning_rate, dropout, initialisation, 227, 3);
add_convolution(network, 11, 96, 4, 0, activation);
add_average_pooling(network, 3, 2, 0);
add_convolution(network, 5, 256, 1, 2, activation);
add_average_pooling(network, 3, 2, 0);
add_convolution(network, 3, 384, 1, 1, activation);
add_convolution(network, 3, 384, 1, 1, activation);
add_convolution(network, 3, 256, 1, 1, activation);
add_average_pooling(network, 3, 2, 0);
add_dense_linearisation(network, 4096, activation);
add_dense(network, 4096, activation);
add_dense(network, size_output, SOFTMAX);
return network;
}
Network* create_network_VGG16(float learning_rate, int dropout, int activation, int initialisation, int size_output) {
Network* network = create_network(23, learning_rate, dropout, initialisation, 256, 3);
add_convolution(network, 3, 64, 1, 0, activation); // Conv3-64
add_convolution(network, 3, 64, 1, 0, activation); // Conv3-64
add_average_pooling(network, 2, 2, 0); // Max Pool
add_convolution(network, 3, 128, 1, 0, activation); // Conv3-128
add_convolution(network, 1, 128, 1, 0, activation); // Conv1-128
add_average_pooling(network, 2, 2, 0); // Max Pool
add_convolution(network, 3, 256, 1, 0, activation); // Conv3-256
add_convolution(network, 3, 256, 1, 0, activation); // Conv3-256
add_convolution(network, 1, 256, 1, 0, activation); // Conv1-256
add_average_pooling(network, 2, 2, 0); // Max Pool
add_convolution(network, 3, 512, 1, 0, activation); // Conv3-512
add_convolution(network, 3, 512, 1, 0, activation); // Conv3-512
add_convolution(network, 1, 512, 1, 0, activation); // Conv1-512
add_average_pooling(network, 2, 2, 0); // Max Pool
add_convolution(network, 3, 512, 1, 0, activation); // Conv3-512
add_convolution(network, 3, 512, 1, 0, activation); // Conv3-512
add_convolution(network, 1, 512, 1, 0, activation); // Conv1-512
add_average_pooling(network, 2, 2, 0); // Max Pool
add_dense_linearisation(network, 2048, activation);
add_dense(network, 2048, activation);
add_dense(network, 256, activation);
add_dense(network, size_output, SOFTMAX);
return network;
}
Network* create_simple_one(float learning_rate, int dropout, int activation, int initialisation, int input_width, int input_depth) {
Network* network = create_network(3, learning_rate, dropout, initialisation, input_width, input_depth);
add_dense_linearisation(network, 80, activation);
add_dense(network, 10, SOFTMAX);
return network;
}

20
src/cnn/neuron_io.c
View File

@ -10,8 +10,7 @@
#include "include/neuron_io.h"
#define INITIAL_MAGIC_NUMBER 1010
#define MAGIC_NUMBER 1013 // Increment this whenever you change the code
#define MAGIC_NUMBER 1012
#define CNN 0
#define NN 1
@ -115,11 +114,13 @@ void write_couche(Network* network, int indice_couche, int type_couche, FILE* pt
Kernel_nn* nn = kernel->nn;
// Écriture du pré-corps
uint32_t pre_buffer[4];
uint32_t pre_buffer[6];
pre_buffer[0] = kernel->activation;
pre_buffer[1] = kernel->linearisation;
pre_buffer[2] = nn->size_input;
pre_buffer[3] = nn->size_output;
pre_buffer[4] = kernel->stride;
pre_buffer[5] = kernel->padding;
fwrite(pre_buffer, sizeof(pre_buffer), 1, ptr);
// Écriture du corps
@ -162,12 +163,7 @@ Network* read_network(char* filename) {
(void) !fread(&magic, sizeof(uint32_t), 1, ptr);
if (magic != MAGIC_NUMBER) {
printf_error((char*)"Incorrect magic number !\n");
if (INITIAL_MAGIC_NUMBER < magic && magic >= INITIAL_MAGIC_NUMBER) {
printf("\tThis backup is no longer supported\n");
printf("\tnPlease update it manually or re-train the network.\n");
printf("\t(You can update it with a script or manually with a Hex Editor)\n");
}
printf_error("Incorrect magic number !\n");
exit(1);
}
@ -334,15 +330,15 @@ Kernel* read_kernel(int type_couche, int output_width, FILE* ptr) {
// Lecture du "Pré-corps"
kernel->nn = (Kernel_nn*)nalloc(1, sizeof(Kernel_nn));
kernel->cnn = NULL;
uint32_t buffer[4];
uint32_t buffer[6];
(void) !fread(&buffer, sizeof(buffer), 1, ptr);
kernel->activation = buffer[0];
kernel->linearisation = buffer[1];
kernel->nn->size_input = buffer[2];
kernel->nn->size_output = buffer[3];
kernel->padding = -1;
kernel->stride = -1;
kernel->stride = buffer[4];
kernel->padding = buffer[5];
// Lecture du corps
Kernel_nn* nn = kernel->nn;

6
src/cnn/train.c
View File

@ -8,15 +8,15 @@
#include <omp.h>
#include "../common/include/memory_management.h"
#include "../common/include/colors.h"
#include "../common/include/utils.h"
#include "../common/include/mnist.h"
#include "include/initialisation.h"
#include "include/test_network.h"
#include "include/neuron_io.h"
#include "../common/include/colors.h"
#include "../common/include/utils.h"
#include "include/function.h"
#include "include/creation.h"
#include "include/update.h"
#include "include/models.h"
#include "include/utils.h"
#include "include/free.h"
#include "include/jpeg.h"

1
src/common/memory_management.c
View File

@ -214,7 +214,6 @@ void gree(void* ptr, bool already_freed) {
memory = free_memory(ptr, memory, already_freed);
pthread_mutex_unlock(&memory_lock);
#else
(void)already_freed;
free(ptr);
#endif
}

1
src/common/memory_management.cu
View File

@ -214,7 +214,6 @@ void gree(void* ptr, bool already_freed) {
memory = free_memory(ptr, memory, already_freed);
pthread_mutex_unlock(&memory_lock);
#else
(void)already_freed;
free(ptr);
#endif
}

1
test/cnn_neuron_io.c
View File

@ -6,7 +6,6 @@
#include "../src/common/include/colors.h"
#include "../src/cnn/include/neuron_io.h"
#include "../src/cnn/include/creation.h"
#include "../src/cnn/include/models.h"
#include "../src/cnn/include/utils.h"
#include "../src/cnn/include/free.h"

1
test/cnn_structure.c
View File

@ -5,7 +5,6 @@
#include "../src/common/include/colors.h"
#include "../src/cnn/include/creation.h"
#include "../src/cnn/include/models.h"
#include "../src/cnn/include/utils.h"
#include "../src/cnn/include/free.h"

1
test/cnn_utils.c
View File

@ -3,7 +3,6 @@
#include "../src/common/include/colors.h"
#include "../src/cnn/include/creation.h"
#include "../src/cnn/include/models.h"
#include "../src/cnn/include/utils.h"
#include "../src/cnn/include/free.h"