From 0f1a31648c5292fa49b35eac90a2ee676d6c13e6 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Sat, 31 Jan 2015 06:05:23 +0000
Subject: [PATCH] idk, probably something changed
---
src/convolutional_layer.c | 131 ++++++++++++++++++-------------------------
1 files changed, 54 insertions(+), 77 deletions(-)
diff --git a/src/convolutional_layer.c b/src/convolutional_layer.c
index 44e9244..6848511 100644
--- a/src/convolutional_layer.c
+++ b/src/convolutional_layer.c
@@ -1,7 +1,11 @@
#include "convolutional_layer.h"
#include "utils.h"
-#include "mini_blas.h"
+#include "im2col.h"
+#include "col2im.h"
+#include "blas.h"
+#include "gemm.h"
#include <stdio.h>
+#include <time.h>
int convolutional_out_height(convolutional_layer layer)
{
@@ -37,11 +41,16 @@
return float_to_image(h,w,c,layer.delta);
}
-convolutional_layer *make_convolutional_layer(int batch, int h, int w, int c, int n, int size, int stride, int pad, ACTIVATION activation)
+convolutional_layer *make_convolutional_layer(int batch, int h, int w, int c, int n, int size, int stride, int pad, ACTIVATION activation, float learning_rate, float momentum, float decay)
{
int i;
size = 2*(size/2)+1; //HA! And you thought you'd use an even sized filter...
convolutional_layer *layer = calloc(1, sizeof(convolutional_layer));
+
+ layer->learning_rate = learning_rate;
+ layer->momentum = momentum;
+ layer->decay = decay;
+
layer->h = h;
layer->w = w;
layer->c = c;
@@ -53,35 +62,33 @@
layer->filters = calloc(c*n*size*size, sizeof(float));
layer->filter_updates = calloc(c*n*size*size, sizeof(float));
- layer->filter_momentum = calloc(c*n*size*size, sizeof(float));
layer->biases = calloc(n, sizeof(float));
layer->bias_updates = calloc(n, sizeof(float));
- layer->bias_momentum = calloc(n, sizeof(float));
- float scale = 1./(size*size*c);
- for(i = 0; i < c*n*size*size; ++i) layer->filters[i] = scale*(rand_uniform());
+ float scale = 1./sqrt(size*size*c);
+ //scale = .05;
+ for(i = 0; i < c*n*size*size; ++i) layer->filters[i] = scale*rand_normal();
for(i = 0; i < n; ++i){
//layer->biases[i] = rand_normal()*scale + scale;
- layer->biases[i] = .5;
+ layer->biases[i] = scale;
}
int out_h = convolutional_out_height(*layer);
int out_w = convolutional_out_width(*layer);
- layer->col_image = calloc(layer->batch*out_h*out_w*size*size*c, sizeof(float));
+ layer->col_image = calloc(out_h*out_w*size*size*c, sizeof(float));
layer->output = calloc(layer->batch*out_h * out_w * n, sizeof(float));
layer->delta = calloc(layer->batch*out_h * out_w * n, sizeof(float));
+
#ifdef GPU
- layer->filters_cl = cl_make_array(layer->filters, c*n*size*size);
- layer->filter_updates_cl = cl_make_array(layer->filter_updates, c*n*size*size);
- layer->filter_momentum_cl = cl_make_array(layer->filter_momentum, c*n*size*size);
+ layer->filters_gpu = cuda_make_array(layer->filters, c*n*size*size);
+ layer->filter_updates_gpu = cuda_make_array(layer->filter_updates, c*n*size*size);
- layer->biases_cl = cl_make_array(layer->biases, n);
- layer->bias_updates_cl = cl_make_array(layer->bias_updates, n);
- layer->bias_momentum_cl = cl_make_array(layer->bias_momentum, n);
+ layer->biases_gpu = cuda_make_array(layer->biases, n);
+ layer->bias_updates_gpu = cuda_make_array(layer->bias_updates, n);
- layer->col_image_cl = cl_make_array(layer->col_image, layer.batch*out_h*out_w*size*size*c);
- layer->delta_cl = cl_make_array(layer->delta, layer->batch*out_h*out_w*n);
- layer->output_cl = cl_make_array(layer->output, layer->batch*out_h*out_w*n);
+ layer->col_image_gpu = cuda_make_array(layer->col_image, out_h*out_w*size*size*c);
+ layer->delta_gpu = cuda_make_array(layer->delta, layer->batch*out_h*out_w*n);
+ layer->output_gpu = cuda_make_array(layer->output, layer->batch*out_h*out_w*n);
#endif
layer->activation = activation;
@@ -99,7 +106,7 @@
int out_w = convolutional_out_width(*layer);
layer->col_image = realloc(layer->col_image,
- layer->batch*out_h*out_w*layer->size*layer->size*layer->c*sizeof(float));
+ out_h*out_w*layer->size*layer->size*layer->c*sizeof(float));
layer->output = realloc(layer->output,
layer->batch*out_h * out_w * layer->n*sizeof(float));
layer->delta = realloc(layer->delta,
@@ -141,37 +148,11 @@
layer.size, layer.stride, layer.pad, b);
gemm(0,0,m,n,k,1,a,k,b,n,1,c,n);
c += n*m;
- in += layer.h*layer.w*layer.c;
- b += k*n;
+ in += layer.c*layer.h*layer.w;
}
- /*
- int i;
- for(i = 0; i < m*n; ++i) printf("%f, ", layer.output[i]);
- printf("\n");
- */
- activate_array(layer.output, m*n*layer.batch, layer.activation, 0.);
+ activate_array(layer.output, m*n*layer.batch, layer.activation);
}
-#ifdef GPU
-void forward_convolutional_layer_gpu(convolutional_layer layer, cl_mem in)
-{
- int m = layer.n;
- int k = layer.size*layer.size*layer.c;
- int n = convolutional_out_height(layer)*
- convolutional_out_width(layer)*
- layer.batch;
-
- cl_write_array(layer.filters_cl, layer.filters, m*k);
- cl_mem a = layer.filters_cl;
- cl_mem b = layer.col_image_cl;
- cl_mem c = layer.output_cl;
- im2col_ongpu(in, layer.batch, layer.c, layer.h, layer.w, layer.size, layer.stride, b);
- gemm_ongpu(0,0,m,n,k,1,a,k,b,n,0,c,n);
- activate_array_ongpu(layer.output_cl, m*n, layer.activation, 0.);
- cl_read_array(layer.output_cl, layer.output, m*n);
-}
-#endif
-
void learn_bias_convolutional_layer(convolutional_layer layer)
{
int i,b;
@@ -179,61 +160,57 @@
*convolutional_out_width(layer);
for(b = 0; b < layer.batch; ++b){
for(i = 0; i < layer.n; ++i){
- layer.bias_updates[i] += mean_array(layer.delta+size*(i+b*layer.n), size);
+ layer.bias_updates[i] += sum_array(layer.delta+size*(i+b*layer.n), size);
}
}
}
-void backward_convolutional_layer(convolutional_layer layer, float *delta)
+void backward_convolutional_layer(convolutional_layer layer, float *in, float *delta)
{
int i;
int m = layer.n;
int n = layer.size*layer.size*layer.c;
int k = convolutional_out_height(layer)*
convolutional_out_width(layer);
+
gradient_array(layer.output, m*k*layer.batch, layer.activation, layer.delta);
+
learn_bias_convolutional_layer(layer);
- float *a = layer.delta;
- float *b = layer.col_image;
- float *c = layer.filter_updates;
+ if(delta) memset(delta, 0, layer.batch*layer.h*layer.w*layer.c*sizeof(float));
for(i = 0; i < layer.batch; ++i){
+ float *a = layer.delta + i*m*k;
+ float *b = layer.col_image;
+ float *c = layer.filter_updates;
+
+ float *im = in+i*layer.c*layer.h*layer.w;
+
+ im2col_cpu(im, layer.c, layer.h, layer.w,
+ layer.size, layer.stride, layer.pad, b);
gemm(0,1,m,n,k,1,a,k,b,k,1,c,n);
- a += m*k;
- b += k*n;
- }
- if(delta){
- m = layer.size*layer.size*layer.c;
- k = layer.n;
- n = convolutional_out_height(layer)*
- convolutional_out_width(layer);
+ if(delta){
+ a = layer.filters;
+ b = layer.delta + i*m*k;
+ c = layer.col_image;
- a = layer.filters;
- b = layer.delta;
- c = layer.col_image;
+ gemm(1,0,n,k,m,1,a,n,b,k,0,c,k);
- memset(delta, 0, layer.batch*layer.h*layer.w*layer.c*sizeof(float));
-
- for(i = 0; i < layer.batch; ++i){
- gemm(1,0,m,n,k,1,a,m,b,n,0,c,n);
- col2im_cpu(c, layer.c, layer.h, layer.w, layer.size, layer.stride, layer.pad, delta);
- c += k*n;
- delta += layer.h*layer.w*layer.c;
+ col2im_cpu(layer.col_image, layer.c, layer.h, layer.w, layer.size, layer.stride, layer.pad, delta+i*layer.c*layer.h*layer.w);
}
}
}
-void update_convolutional_layer(convolutional_layer layer, float step, float momentum, float decay)
+void update_convolutional_layer(convolutional_layer layer)
{
int size = layer.size*layer.size*layer.c*layer.n;
- axpy_cpu(layer.n, step, layer.bias_updates, 1, layer.biases, 1);
- scal_cpu(layer.n, momentum, layer.bias_updates, 1);
+ axpy_cpu(layer.n, layer.learning_rate, layer.bias_updates, 1, layer.biases, 1);
+ scal_cpu(layer.n, layer.momentum, layer.bias_updates, 1);
- scal_cpu(size, 1.-step*decay, layer.filters, 1);
- axpy_cpu(size, step, layer.filter_updates, 1, layer.filters, 1);
- scal_cpu(size, momentum, layer.filter_updates, 1);
+ axpy_cpu(size, -layer.decay, layer.filters, 1, layer.filter_updates, 1);
+ axpy_cpu(size, layer.learning_rate, layer.filter_updates, 1, layer.filters, 1);
+ scal_cpu(size, layer.momentum, layer.filter_updates, 1);
}
@@ -284,8 +261,8 @@
image dc = collapse_image_layers(delta, 1);
char buff[256];
sprintf(buff, "%s: Output", window);
- show_image(dc, buff);
- save_image(dc, buff);
+ //show_image(dc, buff);
+ //save_image(dc, buff);
free_image(dc);
return single_filters;
}
--
Gitblit v1.10.0