From 787d5345609459f21fd65d2d8b4fcd55201e21a1 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Mon, 13 Oct 2014 07:31:10 +0000
Subject: [PATCH] Convolutional working on GPU
---
src/maxpool_layer.c | 116 +++++++++++++++++----------------------------------------
1 files changed, 35 insertions(+), 81 deletions(-)
diff --git a/src/maxpool_layer.c b/src/maxpool_layer.c
index 08c9f2f..070eaba 100644
--- a/src/maxpool_layer.c
+++ b/src/maxpool_layer.c
@@ -27,6 +27,7 @@
layer->c = c;
layer->size = size;
layer->stride = stride;
+ layer->max_indexes = calloc(((h-1)/stride+1) * ((w-1)/stride+1) * c*batch, sizeof(int));
layer->output = calloc(((h-1)/stride+1) * ((w-1)/stride+1) * c*batch, sizeof(float));
layer->delta = calloc(((h-1)/stride+1) * ((w-1)/stride+1) * c*batch, sizeof(float));
return layer;
@@ -41,101 +42,54 @@
layer->delta = realloc(layer->delta, ((h-1)/layer->stride+1) * ((w-1)/layer->stride+1) * c * layer->batch*sizeof(float));
}
-float get_max_region(image im, int h, int w, int c, int size)
-{
- int i,j;
- int lower = (-size-1)/2 + 1;
- int upper = size/2 + 1;
-
- int lh = (h-lower < 0) ? 0 : h-lower;
- int uh = (h+upper > im.h) ? im.h : h+upper;
-
- int lw = (w-lower < 0) ? 0 : w-lower;
- int uw = (w+upper > im.w) ? im.w : w+upper;
-
- //printf("%d\n", -3/2);
- //printf("%d %d\n", lower, upper);
- //printf("%d %d %d %d\n", lh, uh, lw, uw);
-
- float max = -FLT_MAX;
- for(i = lh; i < uh; ++i){
- for(j = lw; j < uw; ++j){
- float val = get_pixel(im, i, j, c);
- if (val > max) max = val;
- }
- }
- return max;
-}
-
-void forward_maxpool_layer(const maxpool_layer layer, float *in)
+void forward_maxpool_layer(const maxpool_layer layer, float *input)
{
int b;
for(b = 0; b < layer.batch; ++b){
- image input = float_to_image(layer.h, layer.w, layer.c, in+b*layer.h*layer.w*layer.c);
-
int h = (layer.h-1)/layer.stride + 1;
int w = (layer.w-1)/layer.stride + 1;
int c = layer.c;
- image output = float_to_image(h,w,c,layer.output+b*h*w*c);
- int i,j,k;
- for(k = 0; k < input.c; ++k){
- for(i = 0; i < input.h; i += layer.stride){
- for(j = 0; j < input.w; j += layer.stride){
- float max = get_max_region(input, i, j, k, layer.size);
- set_pixel(output, i/layer.stride, j/layer.stride, k, max);
+ int i,j,k,l,m;
+ for(k = 0; k < layer.c; ++k){
+ for(i = 0; i < layer.h; i += layer.stride){
+ for(j = 0; j < layer.w; j += layer.stride){
+ int out_index = j/layer.stride + w*(i/layer.stride + h*(k + c*b));
+ layer.output[out_index] = -FLT_MAX;
+ int lower = (-layer.size-1)/2 + 1;
+ int upper = layer.size/2 + 1;
+
+ int lh = (i+lower < 0) ? 0 : i+lower;
+ int uh = (i+upper > layer.h) ? layer.h : i+upper;
+
+ int lw = (j+lower < 0) ? 0 : j+lower;
+ int uw = (j+upper > layer.w) ? layer.w : j+upper;
+ for(l = lh; l < uh; ++l){
+ for(m = lw; m < uw; ++m){
+ //printf("%d %d\n", l, m);
+ int index = m + layer.w*(l + layer.h*(k + b*layer.c));
+ if(input[index] > layer.output[out_index]){
+ layer.output[out_index] = input[index];
+ layer.max_indexes[out_index] = index;
+ }
+ }
+ }
}
}
}
}
}
-float set_max_region_delta(image im, image delta, int h, int w, int c, int size, float max, float error)
+void backward_maxpool_layer(const maxpool_layer layer, float *input, float *delta)
{
- int i,j;
- int lower = (-size-1)/2 + 1;
- int upper = size/2 + 1;
-
- int lh = (h-lower < 0) ? 0 : h-lower;
- int uh = (h+upper > im.h) ? im.h : h+upper;
-
- int lw = (w-lower < 0) ? 0 : w-lower;
- int uw = (w+upper > im.w) ? im.w : w+upper;
-
- for(i = lh; i < uh; ++i){
- for(j = lw; j < uw; ++j){
- float val = get_pixel(im, i, j, c);
- if (val == max){
- add_pixel(delta, i, j, c, error);
- }
- }
- }
- return max;
-}
-
-void backward_maxpool_layer(const maxpool_layer layer, float *in, float *delta)
-{
- int b;
- for(b = 0; b < layer.batch; ++b){
- image input = float_to_image(layer.h, layer.w, layer.c, in+b*layer.h*layer.w*layer.c);
- image input_delta = float_to_image(layer.h, layer.w, layer.c, delta+b*layer.h*layer.w*layer.c);
- int h = (layer.h-1)/layer.stride + 1;
- int w = (layer.w-1)/layer.stride + 1;
- int c = layer.c;
- image output = float_to_image(h,w,c,layer.output+b*h*w*c);
- image output_delta = float_to_image(h,w,c,layer.delta+b*h*w*c);
- zero_image(input_delta);
-
- int i,j,k;
- for(k = 0; k < input.c; ++k){
- for(i = 0; i < input.h; i += layer.stride){
- for(j = 0; j < input.w; j += layer.stride){
- float max = get_pixel(output, i/layer.stride, j/layer.stride, k);
- float error = get_pixel(output_delta, i/layer.stride, j/layer.stride, k);
- set_max_region_delta(input, input_delta, i, j, k, layer.size, max, error);
- }
- }
- }
+ int i;
+ int h = (layer.h-1)/layer.stride + 1;
+ int w = (layer.w-1)/layer.stride + 1;
+ int c = layer.c;
+ memset(delta, 0, layer.batch*layer.h*layer.w*layer.c*sizeof(float));
+ for(i = 0; i < h*w*c*layer.batch; ++i){
+ int index = layer.max_indexes[i];
+ delta[index] += layer.delta[i];
}
}
--
Gitblit v1.10.0