mirror of
https://github.com/pjreddie/darknet.git
synced 2023-08-10 21:13:14 +03:00
First Commit!
This commit is contained in:
commit
41bcfac86f
13
.gitignore
vendored
Normal file
13
.gitignore
vendored
Normal file
@ -0,0 +1,13 @@
|
||||
*.o
|
||||
*.dSYM
|
||||
*.csv
|
||||
images/
|
||||
opencv/
|
||||
cnn
|
||||
|
||||
# OS Generated #
|
||||
.DS_Store*
|
||||
ehthumbs.db
|
||||
Icon?
|
||||
Thumbs.db
|
||||
*.swp
|
21
Makefile
Normal file
21
Makefile
Normal file
@ -0,0 +1,21 @@
|
||||
CC=gcc
|
||||
CFLAGS=-Wall `pkg-config --cflags opencv` -O3 -flto -ffast-math
|
||||
#CFLAGS=-Wall `pkg-config --cflags opencv` -O0 -g
|
||||
LDFLAGS=`pkg-config --libs opencv` -lm
|
||||
VPATH=./src/
|
||||
|
||||
OBJ=network.o image.o tests.o convolutional_layer.o connected_layer.o maxpool_layer.o
|
||||
|
||||
all: cnn
|
||||
|
||||
cnn: $(OBJ)
|
||||
$(CC) $(CFLAGS) $(LDFLAGS) $^ -o $@
|
||||
|
||||
%.o: %.c
|
||||
$(CC) $(CFLAGS) -c $< -o $@
|
||||
|
||||
.PHONY: clean
|
||||
|
||||
clean:
|
||||
rm -rf $(OBJ) cnn
|
||||
|
92
src/connected_layer.c
Normal file
92
src/connected_layer.c
Normal file
@ -0,0 +1,92 @@
|
||||
#include "connected_layer.h"
|
||||
|
||||
#include <stdlib.h>
|
||||
#include <string.h>
|
||||
|
||||
double activation(double x)
|
||||
{
|
||||
return x*(x>0);
|
||||
}
|
||||
|
||||
double gradient(double x)
|
||||
{
|
||||
return (x>=0);
|
||||
}
|
||||
|
||||
connected_layer make_connected_layer(int inputs, int outputs)
|
||||
{
|
||||
int i;
|
||||
connected_layer layer;
|
||||
layer.inputs = inputs;
|
||||
layer.outputs = outputs;
|
||||
|
||||
layer.output = calloc(outputs, sizeof(double*));
|
||||
|
||||
layer.weight_updates = calloc(inputs*outputs, sizeof(double));
|
||||
layer.weights = calloc(inputs*outputs, sizeof(double));
|
||||
for(i = 0; i < inputs*outputs; ++i)
|
||||
layer.weights[i] = .5 - (double)rand()/RAND_MAX;
|
||||
|
||||
layer.bias_updates = calloc(outputs, sizeof(double));
|
||||
layer.biases = calloc(outputs, sizeof(double));
|
||||
for(i = 0; i < outputs; ++i)
|
||||
layer.biases[i] = (double)rand()/RAND_MAX;
|
||||
|
||||
return layer;
|
||||
}
|
||||
|
||||
void run_connected_layer(double *input, connected_layer layer)
|
||||
{
|
||||
int i, j;
|
||||
for(i = 0; i < layer.outputs; ++i){
|
||||
layer.output[i] = layer.biases[i];
|
||||
for(j = 0; j < layer.inputs; ++j){
|
||||
layer.output[i] += input[j]*layer.weights[i*layer.outputs + j];
|
||||
}
|
||||
layer.output[i] = activation(layer.output[i]);
|
||||
}
|
||||
}
|
||||
|
||||
void backpropagate_connected_layer(double *input, connected_layer layer)
|
||||
{
|
||||
int i, j;
|
||||
double *old_input = calloc(layer.inputs, sizeof(double));
|
||||
memcpy(old_input, input, layer.inputs*sizeof(double));
|
||||
memset(input, 0, layer.inputs*sizeof(double));
|
||||
|
||||
for(i = 0; i < layer.outputs; ++i){
|
||||
for(j = 0; j < layer.inputs; ++j){
|
||||
input[j] += layer.output[i]*layer.weights[i*layer.outputs + j];
|
||||
}
|
||||
}
|
||||
for(j = 0; j < layer.inputs; ++j){
|
||||
input[j] = input[j]*gradient(old_input[j]);
|
||||
}
|
||||
free(old_input);
|
||||
}
|
||||
|
||||
void calculate_updates_connected_layer(double *input, connected_layer layer)
|
||||
{
|
||||
int i, j;
|
||||
for(i = 0; i < layer.outputs; ++i){
|
||||
layer.bias_updates[i] += layer.output[i];
|
||||
for(j = 0; j < layer.inputs; ++j){
|
||||
layer.weight_updates[i*layer.outputs + j] += layer.output[i]*input[j];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void update_connected_layer(connected_layer layer, double step)
|
||||
{
|
||||
int i,j;
|
||||
for(i = 0; i < layer.outputs; ++i){
|
||||
layer.biases[i] += step*layer.bias_updates[i];
|
||||
for(j = 0; j < layer.inputs; ++j){
|
||||
int index = i*layer.outputs+j;
|
||||
layer.weights[index] = layer.weight_updates[index];
|
||||
}
|
||||
}
|
||||
memset(layer.bias_updates, 0, layer.outputs*sizeof(double));
|
||||
memset(layer.weight_updates, 0, layer.outputs*layer.inputs*sizeof(double));
|
||||
}
|
||||
|
21
src/connected_layer.h
Normal file
21
src/connected_layer.h
Normal file
@ -0,0 +1,21 @@
|
||||
#ifndef CONNECTED_LAYER_H
|
||||
#define CONNECTED_LAYER_H
|
||||
|
||||
typedef struct{
|
||||
int inputs;
|
||||
int outputs;
|
||||
double *weights;
|
||||
double *biases;
|
||||
double *weight_updates;
|
||||
double *bias_updates;
|
||||
double *output;
|
||||
} connected_layer;
|
||||
|
||||
connected_layer make_connected_layer(int inputs, int outputs);
|
||||
void run_connected_layer(double *input, connected_layer layer);
|
||||
void backpropagate_connected_layer(double *input, connected_layer layer);
|
||||
void calculate_updates_connected_layer(double *input, connected_layer layer);
|
||||
void update_connected_layer(connected_layer layer, double step);
|
||||
|
||||
#endif
|
||||
|
86
src/convolutional_layer.c
Normal file
86
src/convolutional_layer.c
Normal file
@ -0,0 +1,86 @@
|
||||
#include "convolutional_layer.h"
|
||||
|
||||
double convolution_activation(double x)
|
||||
{
|
||||
return x*(x>0);
|
||||
}
|
||||
|
||||
double convolution_gradient(double x)
|
||||
{
|
||||
return (x>=0);
|
||||
}
|
||||
|
||||
convolutional_layer make_convolutional_layer(int h, int w, int c, int n, int size, int stride)
|
||||
{
|
||||
int i;
|
||||
convolutional_layer layer;
|
||||
layer.n = n;
|
||||
layer.stride = stride;
|
||||
layer.kernels = calloc(n, sizeof(image));
|
||||
layer.kernel_updates = calloc(n, sizeof(image));
|
||||
for(i = 0; i < n; ++i){
|
||||
layer.kernels[i] = make_random_kernel(size, c);
|
||||
layer.kernel_updates[i] = make_random_kernel(size, c);
|
||||
}
|
||||
layer.output = make_image((h-1)/stride+1, (w-1)/stride+1, n);
|
||||
layer.upsampled = make_image(h,w,n);
|
||||
return layer;
|
||||
}
|
||||
|
||||
void run_convolutional_layer(const image input, const convolutional_layer layer)
|
||||
{
|
||||
int i;
|
||||
for(i = 0; i < layer.n; ++i){
|
||||
convolve(input, layer.kernels[i], layer.stride, i, layer.output);
|
||||
}
|
||||
for(i = 0; i < input.h*input.w*input.c; ++i){
|
||||
input.data[i] = convolution_activation(input.data[i]);
|
||||
}
|
||||
}
|
||||
|
||||
void backpropagate_layer(image input, convolutional_layer layer)
|
||||
{
|
||||
int i;
|
||||
zero_image(input);
|
||||
for(i = 0; i < layer.n; ++i){
|
||||
back_convolve(input, layer.kernels[i], layer.stride, i, layer.output);
|
||||
}
|
||||
}
|
||||
|
||||
void backpropagate_layer_convolve(image input, convolutional_layer layer)
|
||||
{
|
||||
int i,j;
|
||||
for(i = 0; i < layer.n; ++i){
|
||||
rotate_image(layer.kernels[i]);
|
||||
}
|
||||
|
||||
zero_image(input);
|
||||
upsample_image(layer.output, layer.stride, layer.upsampled);
|
||||
for(j = 0; j < input.c; ++j){
|
||||
for(i = 0; i < layer.n; ++i){
|
||||
two_d_convolve(layer.upsampled, i, layer.kernels[i], j, 1, input, j);
|
||||
}
|
||||
}
|
||||
|
||||
for(i = 0; i < layer.n; ++i){
|
||||
rotate_image(layer.kernels[i]);
|
||||
}
|
||||
}
|
||||
|
||||
void error_convolutional_layer(image input, convolutional_layer layer)
|
||||
{
|
||||
int i;
|
||||
for(i = 0; i < layer.n; ++i){
|
||||
kernel_update(input, layer.kernel_updates[i], layer.stride, i, layer.output);
|
||||
}
|
||||
image old_input = copy_image(input);
|
||||
zero_image(input);
|
||||
for(i = 0; i < layer.n; ++i){
|
||||
back_convolve(input, layer.kernels[i], layer.stride, i, layer.output);
|
||||
}
|
||||
for(i = 0; i < input.h*input.w*input.c; ++i){
|
||||
input.data[i] = input.data[i]*convolution_gradient(input.data[i]);
|
||||
}
|
||||
free_image(old_input);
|
||||
}
|
||||
|
21
src/convolutional_layer.h
Normal file
21
src/convolutional_layer.h
Normal file
@ -0,0 +1,21 @@
|
||||
#ifndef CONVOLUTIONAL_LAYER_H
|
||||
#define CONVOLUTIONAL_LAYER_H
|
||||
|
||||
#include "image.h"
|
||||
|
||||
typedef struct {
|
||||
int n;
|
||||
int stride;
|
||||
image *kernels;
|
||||
image *kernel_updates;
|
||||
image upsampled;
|
||||
image output;
|
||||
} convolutional_layer;
|
||||
|
||||
convolutional_layer make_convolutional_layer(int w, int h, int c, int n, int size, int stride);
|
||||
void run_convolutional_layer(const image input, const convolutional_layer layer);
|
||||
void backpropagate_layer(image input, convolutional_layer layer);
|
||||
void backpropagate_layer_convolve(image input, convolutional_layer layer);
|
||||
|
||||
#endif
|
||||
|
348
src/image.c
Normal file
348
src/image.c
Normal file
@ -0,0 +1,348 @@
|
||||
#include "image.h"
|
||||
#include <stdio.h>
|
||||
|
||||
int windows = 0;
|
||||
|
||||
void subtract_image(image a, image b)
|
||||
{
|
||||
int i;
|
||||
for(i = 0; i < a.h*a.w*a.c; ++i) a.data[i] -= b.data[i];
|
||||
}
|
||||
|
||||
void normalize_image(image p)
|
||||
{
|
||||
double *min = calloc(p.c, sizeof(double));
|
||||
double *max = calloc(p.c, sizeof(double));
|
||||
int i,j;
|
||||
for(i = 0; i < p.c; ++i) min[i] = max[i] = p.data[i*p.h*p.w];
|
||||
|
||||
for(j = 0; j < p.c; ++j){
|
||||
for(i = 0; i < p.h*p.w; ++i){
|
||||
double v = p.data[i+j*p.h*p.w];
|
||||
if(v < min[j]) min[j] = v;
|
||||
if(v > max[j]) max[j] = v;
|
||||
}
|
||||
}
|
||||
for(i = 0; i < p.c; ++i){
|
||||
if(max[i] - min[i] < .00001){
|
||||
min[i] = 0;
|
||||
max[i] = 1;
|
||||
}
|
||||
}
|
||||
for(j = 0; j < p.c; ++j){
|
||||
for(i = 0; i < p.w*p.h; ++i){
|
||||
p.data[i+j*p.h*p.w] = (p.data[i+j*p.h*p.w] - min[j])/(max[j]-min[j]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void threshold_image(image p, double t)
|
||||
{
|
||||
int i;
|
||||
for(i = 0; i < p.w*p.h*p.c; ++i){
|
||||
if(p.data[i] < t) p.data[i] = 0;
|
||||
}
|
||||
}
|
||||
|
||||
image copy_image(image p)
|
||||
{
|
||||
image copy = p;
|
||||
copy.data = calloc(p.h*p.w*p.c, sizeof(double));
|
||||
memcpy(copy.data, p.data, p.h*p.w*p.c*sizeof(double));
|
||||
return copy;
|
||||
}
|
||||
|
||||
void show_image(image p, char *name)
|
||||
{
|
||||
int i,j,k;
|
||||
image copy = copy_image(p);
|
||||
normalize_image(copy);
|
||||
|
||||
char buff[256];
|
||||
sprintf(buff, "%s (%d)", name, windows);
|
||||
|
||||
IplImage *disp = cvCreateImage(cvSize(p.w,p.h), IPL_DEPTH_8U, p.c);
|
||||
int step = disp->widthStep;
|
||||
cvNamedWindow(buff, CV_WINDOW_AUTOSIZE);
|
||||
cvMoveWindow(buff, 100*(windows%10) + 200*(windows/10), 100*(windows%10));
|
||||
++windows;
|
||||
for(i = 0; i < p.h; ++i){
|
||||
for(j = 0; j < p.w; ++j){
|
||||
for(k= 0; k < p.c; ++k){
|
||||
disp->imageData[i*step + j*p.c + k] = (unsigned char)(get_pixel(copy,i,j,k)*255);
|
||||
}
|
||||
}
|
||||
}
|
||||
if(disp->height < 100 || disp->width < 100){
|
||||
IplImage *buffer = disp;
|
||||
disp = cvCreateImage(cvSize(100,100*p.h/p.w), buffer->depth, buffer->nChannels);
|
||||
cvResize(buffer, disp, CV_INTER_NN);
|
||||
cvReleaseImage(&buffer);
|
||||
}
|
||||
cvShowImage(buff, disp);
|
||||
cvReleaseImage(&disp);
|
||||
}
|
||||
|
||||
void show_image_layers(image p, char *name)
|
||||
{
|
||||
int i;
|
||||
char buff[256];
|
||||
for(i = 0; i < p.c; ++i){
|
||||
sprintf(buff, "%s - Layer %d", name, i);
|
||||
image layer = get_image_layer(p, i);
|
||||
show_image(layer, buff);
|
||||
free_image(layer);
|
||||
}
|
||||
}
|
||||
|
||||
image make_image(int h, int w, int c)
|
||||
{
|
||||
image out;
|
||||
out.h = h;
|
||||
out.w = w;
|
||||
out.c = c;
|
||||
out.data = calloc(h*w*c, sizeof(double));
|
||||
return out;
|
||||
}
|
||||
|
||||
void zero_image(image m)
|
||||
{
|
||||
memset(m.data, 0, m.h*m.w*m.c*sizeof(double));
|
||||
}
|
||||
|
||||
void zero_channel(image m, int c)
|
||||
{
|
||||
memset(&(m.data[c*m.h*m.w]), 0, m.h*m.w*sizeof(double));
|
||||
}
|
||||
|
||||
void rotate_image(image m)
|
||||
{
|
||||
int i,j;
|
||||
for(j = 0; j < m.c; ++j){
|
||||
for(i = 0; i < m.h*m.w/2; ++i){
|
||||
double swap = m.data[j*m.h*m.w + i];
|
||||
m.data[j*m.h*m.w + i] = m.data[j*m.h*m.w + (m.h*m.w-1 - i)];
|
||||
m.data[j*m.h*m.w + (m.h*m.w-1 - i)] = swap;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
image make_random_image(int h, int w, int c)
|
||||
{
|
||||
image out = make_image(h,w,c);
|
||||
int i;
|
||||
for(i = 0; i < h*w*c; ++i){
|
||||
out.data[i] = (double)rand()/RAND_MAX;
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
image make_random_kernel(int size, int c)
|
||||
{
|
||||
int pad;
|
||||
if((pad=(size%2==0))) ++size;
|
||||
image out = make_random_image(size,size,c);
|
||||
int i,k;
|
||||
if(pad){
|
||||
for(k = 0; k < out.c; ++k){
|
||||
for(i = 0; i < size; ++i) {
|
||||
set_pixel(out, i, 0, k, 0);
|
||||
set_pixel(out, 0, i, k, 0);
|
||||
}
|
||||
}
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
|
||||
image load_image(char *filename)
|
||||
{
|
||||
IplImage* src = 0;
|
||||
if( (src = cvLoadImage(filename,-1)) == 0 )
|
||||
{
|
||||
printf("Cannot load file image %s\n", filename);
|
||||
exit(0);
|
||||
}
|
||||
unsigned char *data = (unsigned char *)src->imageData;
|
||||
int c = src->nChannels;
|
||||
int h = src->height;
|
||||
int w = src->width;
|
||||
int step = src->widthStep;
|
||||
image out = make_image(h,w,c);
|
||||
int i, j, k, count=0;;
|
||||
|
||||
for(k= 0; k < c; ++k){
|
||||
for(i = 0; i < h; ++i){
|
||||
for(j = 0; j < w; ++j){
|
||||
out.data[count++] = data[i*step + j*c + k];
|
||||
}
|
||||
}
|
||||
}
|
||||
cvReleaseImage(&src);
|
||||
return out;
|
||||
}
|
||||
|
||||
image get_image_layer(image m, int l)
|
||||
{
|
||||
image out = make_image(m.h, m.w, 1);
|
||||
int i;
|
||||
for(i = 0; i < m.h*m.w; ++i){
|
||||
out.data[i] = m.data[i+l*m.h*m.w];
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
double get_pixel(image m, int x, int y, int c)
|
||||
{
|
||||
assert(x < m.h && y < m.w && c < m.c);
|
||||
return m.data[c*m.h*m.w + x*m.w + y];
|
||||
}
|
||||
double get_pixel_extend(image m, int x, int y, int c)
|
||||
{
|
||||
if(x < 0 || x >= m.h || y < 0 || y >= m.w || c < 0 || c >= m.c) return 0;
|
||||
return get_pixel(m, x, y, c);
|
||||
}
|
||||
void set_pixel(image m, int x, int y, int c, double val)
|
||||
{
|
||||
assert(x < m.h && y < m.w && c < m.c);
|
||||
m.data[c*m.h*m.w + x*m.w + y] = val;
|
||||
}
|
||||
void set_pixel_extend(image m, int x, int y, int c, double val)
|
||||
{
|
||||
if(x < 0 || x >= m.h || y < 0 || y >= m.w || c < 0 || c >= m.c) return;
|
||||
set_pixel(m, x, y, c, val);
|
||||
}
|
||||
|
||||
void add_pixel(image m, int x, int y, int c, double val)
|
||||
{
|
||||
assert(x < m.h && y < m.w && c < m.c);
|
||||
m.data[c*m.h*m.w + x*m.w + y] += val;
|
||||
}
|
||||
|
||||
void add_pixel_extend(image m, int x, int y, int c, double val)
|
||||
{
|
||||
if(x < 0 || x >= m.h || y < 0 || y >= m.w || c < 0 || c >= m.c) return;
|
||||
add_pixel(m, x, y, c, val);
|
||||
}
|
||||
|
||||
void two_d_convolve(image m, int mc, image kernel, int kc, int stride, image out, int oc)
|
||||
{
|
||||
int x,y,i,j;
|
||||
for(x = 0; x < m.h; x += stride){
|
||||
for(y = 0; y < m.w; y += stride){
|
||||
double sum = 0;
|
||||
for(i = 0; i < kernel.h; ++i){
|
||||
for(j = 0; j < kernel.w; ++j){
|
||||
sum += get_pixel(kernel, i, j, kc)*get_pixel_extend(m, x+i-kernel.h/2, y+j-kernel.w/2, mc);
|
||||
}
|
||||
}
|
||||
add_pixel(out, x/stride, y/stride, oc, sum);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
double single_convolve(image m, image kernel, int x, int y)
|
||||
{
|
||||
double sum = 0;
|
||||
int i, j, k;
|
||||
for(i = 0; i < kernel.h; ++i){
|
||||
for(j = 0; j < kernel.w; ++j){
|
||||
for(k = 0; k < kernel.c; ++k){
|
||||
sum += get_pixel(kernel, i, j, k)*get_pixel_extend(m, x+i-kernel.h/2, y+j-kernel.w/2, k);
|
||||
}
|
||||
}
|
||||
}
|
||||
return sum;
|
||||
}
|
||||
|
||||
void convolve(image m, image kernel, int stride, int channel, image out)
|
||||
{
|
||||
assert(m.c == kernel.c);
|
||||
int i;
|
||||
zero_channel(out, channel);
|
||||
for(i = 0; i < m.c; ++i){
|
||||
two_d_convolve(m, i, kernel, i, stride, out, channel);
|
||||
}
|
||||
/*
|
||||
int j;
|
||||
for(i = 0; i < m.h; i += stride){
|
||||
for(j = 0; j < m.w; j += stride){
|
||||
double val = single_convolve(m, kernel, i, j);
|
||||
set_pixel(out, i/stride, j/stride, channel, val);
|
||||
}
|
||||
}
|
||||
*/
|
||||
}
|
||||
|
||||
void upsample_image(image m, int stride, image out)
|
||||
{
|
||||
int i,j,k;
|
||||
zero_image(out);
|
||||
for(k = 0; k < m.c; ++k){
|
||||
for(i = 0; i < m.h; ++i){
|
||||
for(j = 0; j< m.w; ++j){
|
||||
double val = get_pixel(m, i, j, k);
|
||||
set_pixel(out, i*stride, j*stride, k, val);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void single_update(image m, image update, int x, int y, double error)
|
||||
{
|
||||
int i, j, k;
|
||||
for(i = 0; i < update.h; ++i){
|
||||
for(j = 0; j < update.w; ++j){
|
||||
for(k = 0; k < update.c; ++k){
|
||||
double val = get_pixel_extend(m, x+i-update.h/2, y+j-update.w/2, k);
|
||||
add_pixel(update, i, j, k, val*error);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void kernel_update(image m, image update, int stride, int channel, image out)
|
||||
{
|
||||
assert(m.c == update.c);
|
||||
zero_image(update);
|
||||
int i, j;
|
||||
for(i = 0; i < m.h; i += stride){
|
||||
for(j = 0; j < m.w; j += stride){
|
||||
double error = get_pixel(out, i/stride, j/stride, channel);
|
||||
single_update(m, update, i, j, error);
|
||||
}
|
||||
}
|
||||
for(i = 0; i < update.h*update.w*update.c; ++i){
|
||||
update.data[i] /= (m.h/stride)*(m.w/stride);
|
||||
}
|
||||
}
|
||||
|
||||
void single_back_convolve(image m, image kernel, int x, int y, double val)
|
||||
{
|
||||
int i, j, k;
|
||||
for(i = 0; i < kernel.h; ++i){
|
||||
for(j = 0; j < kernel.w; ++j){
|
||||
for(k = 0; k < kernel.c; ++k){
|
||||
double pval = get_pixel(kernel, i, j, k) * val;
|
||||
add_pixel_extend(m, x+i-kernel.h/2, y+j-kernel.w/2, k, pval);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void back_convolve(image m, image kernel, int stride, int channel, image out)
|
||||
{
|
||||
assert(m.c == kernel.c);
|
||||
int i, j;
|
||||
for(i = 0; i < m.h; i += stride){
|
||||
for(j = 0; j < m.w; j += stride){
|
||||
double val = get_pixel(out, i/stride, j/stride, channel);
|
||||
single_back_convolve(m, kernel, i, j, val);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void free_image(image m)
|
||||
{
|
||||
free(m.data);
|
||||
}
|
42
src/image.h
Normal file
42
src/image.h
Normal file
@ -0,0 +1,42 @@
|
||||
#ifndef IMAGE_H
|
||||
#define IMAGE_H
|
||||
|
||||
#include "opencv2/highgui/highgui_c.h"
|
||||
#include "opencv2/imgproc/imgproc_c.h"
|
||||
typedef struct {
|
||||
int h;
|
||||
int w;
|
||||
int c;
|
||||
double *data;
|
||||
} image;
|
||||
|
||||
void normalize_image(image p);
|
||||
void threshold_image(image p, double t);
|
||||
void zero_image(image m);
|
||||
void rotate_image(image m);
|
||||
|
||||
void show_image(image p, char *name);
|
||||
void show_image_layers(image p, char *name);
|
||||
|
||||
image make_image(int h, int w, int c);
|
||||
image make_random_image(int h, int w, int c);
|
||||
image make_random_kernel(int size, int c);
|
||||
image copy_image(image p);
|
||||
image load_image(char *filename);
|
||||
|
||||
double get_pixel(image m, int x, int y, int c);
|
||||
double get_pixel_extend(image m, int x, int y, int c);
|
||||
void set_pixel(image m, int x, int y, int c, double val);
|
||||
|
||||
|
||||
image get_image_layer(image m, int l);
|
||||
|
||||
void two_d_convolve(image m, int mc, image kernel, int kc, int stride, image out, int oc);
|
||||
void upsample_image(image m, int stride, image out);
|
||||
void convolve(image m, image kernel, int stride, int channel, image out);
|
||||
void back_convolve(image m, image kernel, int stride, int channel, image out);
|
||||
void kernel_update(image m, image update, int stride, int channel, image out);
|
||||
|
||||
void free_image(image m);
|
||||
#endif
|
||||
|
24
src/maxpool_layer.c
Normal file
24
src/maxpool_layer.c
Normal file
@ -0,0 +1,24 @@
|
||||
#include "maxpool_layer.h"
|
||||
|
||||
maxpool_layer make_maxpool_layer(int h, int w, int c, int stride)
|
||||
{
|
||||
maxpool_layer layer;
|
||||
layer.stride = stride;
|
||||
layer.output = make_image((h-1)/stride+1, (w-1)/stride+1, c);
|
||||
return layer;
|
||||
}
|
||||
|
||||
void run_maxpool_layer(const image input, const maxpool_layer layer)
|
||||
{
|
||||
int i,j,k;
|
||||
for(i = 0; i < layer.output.h*layer.output.w*layer.output.c; ++i) layer.output.data[i] = -DBL_MAX;
|
||||
for(i = 0; i < input.h; ++i){
|
||||
for(j = 0; j < input.w; ++j){
|
||||
for(k = 0; k < input.c; ++k){
|
||||
double val = get_pixel(input, i, j, k);
|
||||
double cur = get_pixel(layer.output, i/layer.stride, j/layer.stride, k);
|
||||
if(val > cur) set_pixel(layer.output, i/layer.stride, j/layer.stride, k, val);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
15
src/maxpool_layer.h
Normal file
15
src/maxpool_layer.h
Normal file
@ -0,0 +1,15 @@
|
||||
#ifndef MAXPOOL_LAYER_H
|
||||
#define MAXPOOL_LAYER_H
|
||||
|
||||
#include "image.h"
|
||||
|
||||
typedef struct {
|
||||
int stride;
|
||||
image output;
|
||||
} maxpool_layer;
|
||||
|
||||
maxpool_layer make_maxpool_layer(int h, int w, int c, int stride);
|
||||
void run_maxpool_layer(const image input, const maxpool_layer layer);
|
||||
|
||||
#endif
|
||||
|
48
src/network.c
Normal file
48
src/network.c
Normal file
@ -0,0 +1,48 @@
|
||||
#include "network.h"
|
||||
#include "image.h"
|
||||
|
||||
#include "connected_layer.h"
|
||||
#include "convolutional_layer.h"
|
||||
#include "maxpool_layer.h"
|
||||
|
||||
void run_network(image input, network net)
|
||||
{
|
||||
int i;
|
||||
double *input_d = 0;
|
||||
for(i = 0; i < net.n; ++i){
|
||||
if(net.types[i] == CONVOLUTIONAL){
|
||||
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
|
||||
run_convolutional_layer(input, layer);
|
||||
input = layer.output;
|
||||
input_d = layer.output.data;
|
||||
}
|
||||
else if(net.types[i] == CONNECTED){
|
||||
connected_layer layer = *(connected_layer *)net.layers[i];
|
||||
run_connected_layer(input_d, layer);
|
||||
input_d = layer.output;
|
||||
}
|
||||
else if(net.types[i] == MAXPOOL){
|
||||
maxpool_layer layer = *(maxpool_layer *)net.layers[i];
|
||||
run_maxpool_layer(input, layer);
|
||||
input = layer.output;
|
||||
input_d = layer.output.data;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
image get_network_image(network net)
|
||||
{
|
||||
int i;
|
||||
for(i = net.n-1; i >= 0; --i){
|
||||
if(net.types[i] == CONVOLUTIONAL){
|
||||
convolutional_layer layer = *(convolutional_layer *)net.layers[i];
|
||||
return layer.output;
|
||||
}
|
||||
else if(net.types[i] == MAXPOOL){
|
||||
maxpool_layer layer = *(maxpool_layer *)net.layers[i];
|
||||
return layer.output;
|
||||
}
|
||||
}
|
||||
return make_image(1,1,1);
|
||||
}
|
||||
|
23
src/network.h
Normal file
23
src/network.h
Normal file
@ -0,0 +1,23 @@
|
||||
// Oh boy, why am I about to do this....
|
||||
#ifndef NETWORK_H
|
||||
#define NETWORK_H
|
||||
|
||||
#include "image.h"
|
||||
|
||||
typedef enum {
|
||||
CONVOLUTIONAL,
|
||||
CONNECTED,
|
||||
MAXPOOL
|
||||
} LAYER_TYPE;
|
||||
|
||||
typedef struct {
|
||||
int n;
|
||||
void **layers;
|
||||
LAYER_TYPE *types;
|
||||
} network;
|
||||
|
||||
void run_network(image input, network net);
|
||||
image get_network_image(network net);
|
||||
|
||||
#endif
|
||||
|
200
src/tests.c
Normal file
200
src/tests.c
Normal file
@ -0,0 +1,200 @@
|
||||
#include "connected_layer.h"
|
||||
#include "convolutional_layer.h"
|
||||
#include "maxpool_layer.h"
|
||||
#include "network.h"
|
||||
#include "image.h"
|
||||
|
||||
#include <time.h>
|
||||
#include <stdlib.h>
|
||||
#include <stdio.h>
|
||||
|
||||
void test_convolve()
|
||||
{
|
||||
image dog = load_image("dog.jpg");
|
||||
//show_image_layers(dog, "Dog");
|
||||
printf("dog channels %d\n", dog.c);
|
||||
image kernel = make_random_image(3,3,dog.c);
|
||||
image edge = make_image(dog.h, dog.w, 1);
|
||||
int i;
|
||||
clock_t start = clock(), end;
|
||||
for(i = 0; i < 1000; ++i){
|
||||
convolve(dog, kernel, 1, 0, edge);
|
||||
}
|
||||
end = clock();
|
||||
printf("Convolutions: %lf seconds\n", (double)(end-start)/CLOCKS_PER_SEC);
|
||||
show_image_layers(edge, "Test Convolve");
|
||||
}
|
||||
|
||||
void test_color()
|
||||
{
|
||||
image dog = load_image("test_color.png");
|
||||
show_image_layers(dog, "Test Color");
|
||||
}
|
||||
|
||||
void test_convolutional_layer()
|
||||
{
|
||||
srand(0);
|
||||
image dog = load_image("test_dog.jpg");
|
||||
int i;
|
||||
int n = 5;
|
||||
int stride = 1;
|
||||
int size = 8;
|
||||
convolutional_layer layer = make_convolutional_layer(dog.h, dog.w, dog.c, n, size, stride);
|
||||
char buff[256];
|
||||
for(i = 0; i < n; ++i) {
|
||||
sprintf(buff, "Kernel %d", i);
|
||||
show_image(layer.kernels[i], buff);
|
||||
}
|
||||
run_convolutional_layer(dog, layer);
|
||||
|
||||
maxpool_layer mlayer = make_maxpool_layer(layer.output.h, layer.output.w, layer.output.c, 3);
|
||||
run_maxpool_layer(layer.output,mlayer);
|
||||
|
||||
show_image_layers(mlayer.output, "Test Maxpool Layer");
|
||||
}
|
||||
|
||||
void test_load()
|
||||
{
|
||||
image dog = load_image("dog.jpg");
|
||||
show_image(dog, "Test Load");
|
||||
show_image_layers(dog, "Test Load");
|
||||
}
|
||||
void test_upsample()
|
||||
{
|
||||
image dog = load_image("dog.jpg");
|
||||
int n = 3;
|
||||
image up = make_image(n*dog.h, n*dog.w, dog.c);
|
||||
upsample_image(dog, n, up);
|
||||
show_image(up, "Test Upsample");
|
||||
show_image_layers(up, "Test Upsample");
|
||||
}
|
||||
|
||||
void test_rotate()
|
||||
{
|
||||
int i;
|
||||
image dog = load_image("dog.jpg");
|
||||
clock_t start = clock(), end;
|
||||
for(i = 0; i < 1001; ++i){
|
||||
rotate_image(dog);
|
||||
}
|
||||
end = clock();
|
||||
printf("Rotations: %lf seconds\n", (double)(end-start)/CLOCKS_PER_SEC);
|
||||
show_image(dog, "Test Rotate");
|
||||
|
||||
image random = make_random_image(3,3,3);
|
||||
show_image(random, "Test Rotate Random");
|
||||
rotate_image(random);
|
||||
show_image(random, "Test Rotate Random");
|
||||
rotate_image(random);
|
||||
show_image(random, "Test Rotate Random");
|
||||
}
|
||||
|
||||
void test_network()
|
||||
{
|
||||
network net;
|
||||
net.n = 11;
|
||||
net.layers = calloc(net.n, sizeof(void *));
|
||||
net.types = calloc(net.n, sizeof(LAYER_TYPE));
|
||||
net.types[0] = CONVOLUTIONAL;
|
||||
net.types[1] = MAXPOOL;
|
||||
net.types[2] = CONVOLUTIONAL;
|
||||
net.types[3] = MAXPOOL;
|
||||
net.types[4] = CONVOLUTIONAL;
|
||||
net.types[5] = CONVOLUTIONAL;
|
||||
net.types[6] = CONVOLUTIONAL;
|
||||
net.types[7] = MAXPOOL;
|
||||
net.types[8] = CONNECTED;
|
||||
net.types[9] = CONNECTED;
|
||||
net.types[10] = CONNECTED;
|
||||
|
||||
image dog = load_image("test_hinton.jpg");
|
||||
|
||||
int n = 48;
|
||||
int stride = 4;
|
||||
int size = 11;
|
||||
convolutional_layer cl = make_convolutional_layer(dog.h, dog.w, dog.c, n, size, stride);
|
||||
maxpool_layer ml = make_maxpool_layer(cl.output.h, cl.output.w, cl.output.c, 2);
|
||||
|
||||
n = 128;
|
||||
size = 5;
|
||||
stride = 1;
|
||||
convolutional_layer cl2 = make_convolutional_layer(ml.output.h, ml.output.w, ml.output.c, n, size, stride);
|
||||
maxpool_layer ml2 = make_maxpool_layer(cl2.output.h, cl2.output.w, cl2.output.c, 2);
|
||||
|
||||
n = 192;
|
||||
size = 3;
|
||||
convolutional_layer cl3 = make_convolutional_layer(ml2.output.h, ml2.output.w, ml2.output.c, n, size, stride);
|
||||
convolutional_layer cl4 = make_convolutional_layer(cl3.output.h, cl3.output.w, cl3.output.c, n, size, stride);
|
||||
n = 128;
|
||||
convolutional_layer cl5 = make_convolutional_layer(cl4.output.h, cl4.output.w, cl4.output.c, n, size, stride);
|
||||
maxpool_layer ml3 = make_maxpool_layer(cl5.output.h, cl5.output.w, cl5.output.c, 4);
|
||||
connected_layer nl = make_connected_layer(ml3.output.h*ml3.output.w*ml3.output.c, 4096);
|
||||
connected_layer nl2 = make_connected_layer(4096, 4096);
|
||||
connected_layer nl3 = make_connected_layer(4096, 1000);
|
||||
|
||||
net.layers[0] = &cl;
|
||||
net.layers[1] = &ml;
|
||||
net.layers[2] = &cl2;
|
||||
net.layers[3] = &ml2;
|
||||
net.layers[4] = &cl3;
|
||||
net.layers[5] = &cl4;
|
||||
net.layers[6] = &cl5;
|
||||
net.layers[7] = &ml3;
|
||||
net.layers[8] = &nl;
|
||||
net.layers[9] = &nl2;
|
||||
net.layers[10] = &nl3;
|
||||
|
||||
int i;
|
||||
clock_t start = clock(), end;
|
||||
for(i = 0; i < 10; ++i){
|
||||
run_network(dog, net);
|
||||
rotate_image(dog);
|
||||
}
|
||||
end = clock();
|
||||
printf("Ran %lf second per iteration\n", (double)(end-start)/CLOCKS_PER_SEC/10);
|
||||
|
||||
show_image_layers(get_network_image(net), "Test Network Layer");
|
||||
}
|
||||
void test_backpropagate()
|
||||
{
|
||||
int n = 3;
|
||||
int size = 4;
|
||||
int stride = 10;
|
||||
image dog = load_image("dog.jpg");
|
||||
show_image(dog, "Test Backpropagate Input");
|
||||
image dog_copy = copy_image(dog);
|
||||
convolutional_layer cl = make_convolutional_layer(dog.h, dog.w, dog.c, n, size, stride);
|
||||
run_convolutional_layer(dog, cl);
|
||||
show_image(cl.output, "Test Backpropagate Output");
|
||||
int i;
|
||||
clock_t start = clock(), end;
|
||||
for(i = 0; i < 100; ++i){
|
||||
backpropagate_layer(dog_copy, cl);
|
||||
}
|
||||
end = clock();
|
||||
printf("Backpropagate: %lf seconds\n", (double)(end-start)/CLOCKS_PER_SEC);
|
||||
start = clock();
|
||||
for(i = 0; i < 100; ++i){
|
||||
backpropagate_layer_convolve(dog, cl);
|
||||
}
|
||||
end = clock();
|
||||
printf("Backpropagate Using Convolutions: %lf seconds\n", (double)(end-start)/CLOCKS_PER_SEC);
|
||||
show_image(dog_copy, "Test Backpropagate 1");
|
||||
show_image(dog, "Test Backpropagate 2");
|
||||
subtract_image(dog, dog_copy);
|
||||
show_image(dog, "Test Backpropagate Difference");
|
||||
}
|
||||
|
||||
int main()
|
||||
{
|
||||
//test_backpropagate();
|
||||
//test_convolve();
|
||||
//test_upsample();
|
||||
//test_rotate();
|
||||
//test_load();
|
||||
test_network();
|
||||
//test_convolutional_layer();
|
||||
//test_color();
|
||||
cvWaitKey(0);
|
||||
return 0;
|
||||
}
|
BIN
test_color.png
Normal file
BIN
test_color.png
Normal file
Binary file not shown.
After Width: | Height: | Size: 1.1 KiB |
BIN
test_dog.jpg
Normal file
BIN
test_dog.jpg
Normal file
Binary file not shown.
After Width: | Height: | Size: 30 KiB |
BIN
test_hinton.jpg
Normal file
BIN
test_hinton.jpg
Normal file
Binary file not shown.
After Width: | Height: | Size: 16 KiB |
Loading…
Reference in New Issue
Block a user