From 028696bf15efeca3acb3db8c42a96f7b9e0f55ff Mon Sep 17 00:00:00 2001
From: iovodov <b@ovdv.ru>
Date: Thu, 03 May 2018 13:33:46 +0000
Subject: [PATCH] Output improvements for detector results: When printing detector results, output was done in random order, obfuscating results for interpreting. Now: 1. Text output includes coordinates of rects in (left,right,top,bottom in pixels) along with label and score 2. Text output is sorted by rect lefts to simplify finding appropriate rects on image 3. If several class probs are > thresh for some detection, the most probable is written first and coordinates for others are not repeated 4. Rects are imprinted in image in order by their best class prob, so most probable rects are always on top and not overlayed by less probable ones 5. Most probable label for rect is always written first Also: 6. Message about low GPU memory include required amount
---
src/maxpool_layer.c | 212 ++++++++++++++++++++++++----------------------------
1 files changed, 99 insertions(+), 113 deletions(-)
diff --git a/src/maxpool_layer.c b/src/maxpool_layer.c
index 08c9f2f..031d116 100644
--- a/src/maxpool_layer.c
+++ b/src/maxpool_layer.c
@@ -1,141 +1,127 @@
#include "maxpool_layer.h"
+#include "cuda.h"
#include <stdio.h>
-image get_maxpool_image(maxpool_layer layer)
+image get_maxpool_image(maxpool_layer l)
{
- int h = (layer.h-1)/layer.stride + 1;
- int w = (layer.w-1)/layer.stride + 1;
- int c = layer.c;
- return float_to_image(h,w,c,layer.output);
+ int h = l.out_h;
+ int w = l.out_w;
+ int c = l.c;
+ return float_to_image(w,h,c,l.output);
}
-image get_maxpool_delta(maxpool_layer layer)
+image get_maxpool_delta(maxpool_layer l)
{
- int h = (layer.h-1)/layer.stride + 1;
- int w = (layer.w-1)/layer.stride + 1;
- int c = layer.c;
- return float_to_image(h,w,c,layer.delta);
+ int h = l.out_h;
+ int w = l.out_w;
+ int c = l.c;
+ return float_to_image(w,h,c,l.delta);
}
-maxpool_layer *make_maxpool_layer(int batch, int h, int w, int c, int size, int stride)
+maxpool_layer make_maxpool_layer(int batch, int h, int w, int c, int size, int stride, int padding)
{
- fprintf(stderr, "Maxpool Layer: %d x %d x %d image, %d size, %d stride\n", h,w,c,size,stride);
- maxpool_layer *layer = calloc(1, sizeof(maxpool_layer));
- layer->batch = batch;
- layer->h = h;
- layer->w = w;
- layer->c = c;
- layer->size = size;
- layer->stride = stride;
- 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;
+ maxpool_layer l = {0};
+ l.type = MAXPOOL;
+ l.batch = batch;
+ l.h = h;
+ l.w = w;
+ l.c = c;
+ l.pad = padding;
+ l.out_w = (w + 2*padding)/stride;
+ l.out_h = (h + 2*padding)/stride;
+ l.out_c = c;
+ l.outputs = l.out_h * l.out_w * l.out_c;
+ l.inputs = h*w*c;
+ l.size = size;
+ l.stride = stride;
+ int output_size = l.out_h * l.out_w * l.out_c * batch;
+ l.indexes = calloc(output_size, sizeof(int));
+ l.output = calloc(output_size, sizeof(float));
+ l.delta = calloc(output_size, sizeof(float));
+ l.forward = forward_maxpool_layer;
+ l.backward = backward_maxpool_layer;
+ #ifdef GPU
+ l.forward_gpu = forward_maxpool_layer_gpu;
+ l.backward_gpu = backward_maxpool_layer_gpu;
+ l.indexes_gpu = cuda_make_int_array(output_size);
+ l.output_gpu = cuda_make_array(l.output, output_size);
+ l.delta_gpu = cuda_make_array(l.delta, output_size);
+ #endif
+ fprintf(stderr, "max %d x %d / %d %4d x%4d x%4d -> %4d x%4d x%4d\n", size, size, stride, w, h, c, l.out_w, l.out_h, l.out_c);
+ return l;
}
-void resize_maxpool_layer(maxpool_layer *layer, int h, int w, int c)
+void resize_maxpool_layer(maxpool_layer *l, int w, int h)
{
- layer->h = h;
- layer->w = w;
- layer->c = c;
- layer->output = realloc(layer->output, ((h-1)/layer->stride+1) * ((w-1)/layer->stride+1) * c * layer->batch* sizeof(float));
- layer->delta = realloc(layer->delta, ((h-1)/layer->stride+1) * ((w-1)/layer->stride+1) * c * layer->batch*sizeof(float));
+ l->h = h;
+ l->w = w;
+ l->inputs = h*w*l->c;
+
+ l->out_w = (w + 2*l->pad)/l->stride;
+ l->out_h = (h + 2*l->pad)/l->stride;
+ l->outputs = l->out_w * l->out_h * l->c;
+ int output_size = l->outputs * l->batch;
+
+ l->indexes = realloc(l->indexes, output_size * sizeof(int));
+ l->output = realloc(l->output, output_size * sizeof(float));
+ l->delta = realloc(l->delta, output_size * sizeof(float));
+
+ #ifdef GPU
+ cuda_free((float *)l->indexes_gpu);
+ cuda_free(l->output_gpu);
+ cuda_free(l->delta_gpu);
+ l->indexes_gpu = cuda_make_int_array(output_size);
+ l->output_gpu = cuda_make_array(l->output, output_size);
+ l->delta_gpu = cuda_make_array(l->delta, output_size);
+ #endif
}
-float get_max_region(image im, int h, int w, int c, int size)
+void forward_maxpool_layer(const maxpool_layer l, network_state state)
{
- 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 b,i,j,k,m,n;
+ int w_offset = -l.pad;
+ int h_offset = -l.pad;
- 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;
-}
+ int h = l.out_h;
+ int w = l.out_w;
+ int c = l.c;
-void forward_maxpool_layer(const maxpool_layer layer, float *in)
-{
- 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);
+ for(b = 0; b < l.batch; ++b){
+ for(k = 0; k < c; ++k){
+ for(i = 0; i < h; ++i){
+ for(j = 0; j < w; ++j){
+ int out_index = j + w*(i + h*(k + c*b));
+ float max = -FLT_MAX;
+ int max_i = -1;
+ for(n = 0; n < l.size; ++n){
+ for(m = 0; m < l.size; ++m){
+ int cur_h = h_offset + i*l.stride + n;
+ int cur_w = w_offset + j*l.stride + m;
+ int index = cur_w + l.w*(cur_h + l.h*(k + b*l.c));
+ int valid = (cur_h >= 0 && cur_h < l.h &&
+ cur_w >= 0 && cur_w < l.w);
+ float val = (valid != 0) ? state.input[index] : -FLT_MAX;
+ max_i = (val > max) ? index : max_i;
+ max = (val > max) ? val : max;
+ }
+ }
+ l.output[out_index] = max;
+ l.indexes[out_index] = max_i;
}
}
}
}
}
-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 l, network_state state)
{
- 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 = l.out_h;
+ int w = l.out_w;
+ int c = l.c;
+ for(i = 0; i < h*w*c*l.batch; ++i){
+ int index = l.indexes[i];
+ state.delta[index] += l.delta[i];
}
}
--
Gitblit v1.10.0