From 73f7aacf35ec9b1d0f9de9ddf38af0889f213e99 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Tue, 20 Sep 2016 18:34:49 +0000
Subject: [PATCH] better multigpu
---
src/parser.c | 59 +++++++++++++++++++++++++++++++++++++++++++----------------
1 files changed, 43 insertions(+), 16 deletions(-)
diff --git a/src/parser.c b/src/parser.c
index 904df1a..2b285b5 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -2,6 +2,7 @@
#include <string.h>
#include <stdlib.h>
+#include "blas.h"
#include "parser.h"
#include "assert.h"
#include "activations.h"
@@ -147,7 +148,10 @@
int n = option_find_int(options, "filters",1);
int size = option_find_int(options, "size",1);
int stride = option_find_int(options, "stride",1);
- int pad = option_find_int(options, "pad",0);
+ int pad = option_find_int_quiet(options, "pad",0);
+ int padding = option_find_int_quiet(options, "padding",0);
+ if(pad) padding = size/2;
+
char *activation_s = option_find_str(options, "activation", "logistic");
ACTIVATION activation = get_activation(activation_s);
@@ -161,7 +165,7 @@
int binary = option_find_int_quiet(options, "binary", 0);
int xnor = option_find_int_quiet(options, "xnor", 0);
- convolutional_layer layer = make_convolutional_layer(batch,h,w,c,n,size,stride,pad,activation, batch_normalize, binary, xnor);
+ convolutional_layer layer = make_convolutional_layer(batch,h,w,c,n,size,stride,padding,activation, batch_normalize, binary, xnor);
layer.flipped = option_find_int_quiet(options, "flipped", 0);
layer.dot = option_find_float_quiet(options, "dot", 0);
@@ -234,9 +238,16 @@
int coords = option_find_int(options, "coords", 4);
int classes = option_find_int(options, "classes", 20);
int num = option_find_int(options, "num", 1);
+
+ params.w = option_find_int(options, "side", params.w);
+ params.h = option_find_int(options, "side", params.h);
+
layer l = make_region_layer(params.batch, params.w, params.h, num, classes, coords);
assert(l.outputs == params.inputs);
+ l.log = option_find_int_quiet(options, "log", 0);
+ l.sqrt = option_find_int_quiet(options, "sqrt", 0);
+
l.softmax = option_find_int(options, "softmax", 0);
l.max_boxes = option_find_int_quiet(options, "max",30);
l.jitter = option_find_float(options, "jitter", .2);
@@ -278,6 +289,7 @@
COST_TYPE type = get_cost_type(type_s);
float scale = option_find_float_quiet(options, "scale",1);
cost_layer layer = make_cost_layer(params.batch, params.inputs, type, scale);
+ layer.ratio = option_find_float_quiet(options, "ratio",0);
return layer;
}
@@ -324,6 +336,7 @@
{
int stride = option_find_int(options, "stride",1);
int size = option_find_int(options, "size",stride);
+ int padding = option_find_int_quiet(options, "padding", (size-1)/2);
int batch,h,w,c;
h = params.h;
@@ -332,7 +345,7 @@
batch=params.batch;
if(!(h && w && c)) error("Layer before maxpool layer must output image.");
- maxpool_layer layer = make_maxpool_layer(batch,h,w,c,size,stride);
+ maxpool_layer layer = make_maxpool_layer(batch,h,w,c,size,stride,padding);
return layer;
}
@@ -484,8 +497,10 @@
net->min_crop = option_find_int_quiet(options, "min_crop",net->w);
net->angle = option_find_float_quiet(options, "angle", 0);
+ net->aspect = option_find_float_quiet(options, "aspect", 1);
net->saturation = option_find_float_quiet(options, "saturation", 1);
net->exposure = option_find_float_quiet(options, "exposure", 1);
+ net->hue = option_find_float_quiet(options, "hue", 0);
if(!net->inputs && !(net->h && net->w && net->c)) error("No input parameters supplied");
@@ -536,6 +551,7 @@
node *n = sections->front;
if(!n) error("Config file has no sections");
network net = make_network(sections->size - 1);
+ net.gpu_index = gpu_index;
size_params params;
section *s = (section *)n->val;
@@ -841,13 +857,13 @@
for (j = 0; j < l.n; ++j){
int index = j*l.c*l.size*l.size;
- fwrite(l.filters+index, sizeof(float), l.c*l.size*l.size, fp);
+ fwrite(l.weights+index, sizeof(float), l.c*l.size*l.size, fp);
for (k = 0; k < l.c*l.size*l.size; ++k) fwrite(&zero, sizeof(float), 1, fp);
}
for (j = 0; j < l.n; ++j){
int index = j*l.c*l.size*l.size;
for (k = 0; k < l.c*l.size*l.size; ++k) fwrite(&zero, sizeof(float), 1, fp);
- fwrite(l.filters+index, sizeof(float), l.c*l.size*l.size, fp);
+ fwrite(l.weights+index, sizeof(float), l.c*l.size*l.size, fp);
}
}
}
@@ -861,7 +877,7 @@
pull_convolutional_layer(l);
}
#endif
- binarize_filters(l.filters, l.n, l.c*l.size*l.size, l.binary_filters);
+ binarize_weights(l.weights, l.n, l.c*l.size*l.size, l.binary_weights);
int size = l.c*l.size*l.size;
int i, j, k;
fwrite(l.biases, sizeof(float), l.n, fp);
@@ -871,7 +887,7 @@
fwrite(l.rolling_variance, sizeof(float), l.n, fp);
}
for(i = 0; i < l.n; ++i){
- float mean = l.binary_filters[i*size];
+ float mean = l.binary_weights[i*size];
if(mean < 0) mean = -mean;
fwrite(&mean, sizeof(float), 1, fp);
for(j = 0; j < size/8; ++j){
@@ -879,7 +895,7 @@
unsigned char c = 0;
for(k = 0; k < 8; ++k){
if (j*8 + k >= size) break;
- if (l.binary_filters[index + k] > 0) c = (c | 1<<k);
+ if (l.binary_weights[index + k] > 0) c = (c | 1<<k);
}
fwrite(&c, sizeof(char), 1, fp);
}
@@ -904,7 +920,7 @@
fwrite(l.rolling_mean, sizeof(float), l.n, fp);
fwrite(l.rolling_variance, sizeof(float), l.n, fp);
}
- fwrite(l.filters, sizeof(float), num, fp);
+ fwrite(l.weights, sizeof(float), num, fp);
}
void save_batchnorm_weights(layer l, FILE *fp)
@@ -937,6 +953,11 @@
void save_weights_upto(network net, char *filename, int cutoff)
{
+#ifdef GPU
+ if(net.gpu_index >= 0){
+ cuda_set_device(net.gpu_index);
+ }
+#endif
fprintf(stderr, "Saving weights to %s\n", filename);
FILE *fp = fopen(filename, "w");
if(!fp) file_error(filename);
@@ -982,7 +1003,7 @@
int locations = l.out_w*l.out_h;
int size = l.size*l.size*l.c*l.n*locations;
fwrite(l.biases, sizeof(float), l.outputs, fp);
- fwrite(l.filters, sizeof(float), size, fp);
+ fwrite(l.weights, sizeof(float), size, fp);
}
}
fclose(fp);
@@ -1060,7 +1081,7 @@
fread(&c, sizeof(char), 1, fp);
for(k = 0; k < 8; ++k){
if (j*8 + k >= size) break;
- l.filters[index + k] = (c & 1<<k) ? mean : -mean;
+ l.weights[index + k] = (c & 1<<k) ? mean : -mean;
}
}
}
@@ -1084,11 +1105,12 @@
fread(l.rolling_mean, sizeof(float), l.n, fp);
fread(l.rolling_variance, sizeof(float), l.n, fp);
}
- fread(l.filters, sizeof(float), num, fp);
+ fread(l.weights, sizeof(float), num, fp);
+ //if(l.c == 3) scal_cpu(num, 1./256, l.weights, 1);
if (l.flipped) {
- transpose_matrix(l.filters, l.c*l.size*l.size, l.n);
+ transpose_matrix(l.weights, l.c*l.size*l.size, l.n);
}
- //if (l.binary) binarize_filters(l.filters, l.n, l.c*l.size*l.size, l.filters);
+ //if (l.binary) binarize_weights(l.weights, l.n, l.c*l.size*l.size, l.weights);
#ifdef GPU
if(gpu_index >= 0){
push_convolutional_layer(l);
@@ -1099,6 +1121,11 @@
void load_weights_upto(network *net, char *filename, int cutoff)
{
+#ifdef GPU
+ if(net->gpu_index >= 0){
+ cuda_set_device(net->gpu_index);
+ }
+#endif
fprintf(stderr, "Loading weights from %s...", filename);
fflush(stdout);
FILE *fp = fopen(filename, "rb");
@@ -1123,7 +1150,7 @@
if(l.type == DECONVOLUTIONAL){
int num = l.n*l.c*l.size*l.size;
fread(l.biases, sizeof(float), l.n, fp);
- fread(l.filters, sizeof(float), num, fp);
+ fread(l.weights, sizeof(float), num, fp);
#ifdef GPU
if(gpu_index >= 0){
push_deconvolutional_layer(l);
@@ -1158,7 +1185,7 @@
int locations = l.out_w*l.out_h;
int size = l.size*l.size*l.c*l.n*locations;
fread(l.biases, sizeof(float), l.outputs, fp);
- fread(l.filters, sizeof(float), size, fp);
+ fread(l.weights, sizeof(float), size, fp);
#ifdef GPU
if(gpu_index >= 0){
push_local_layer(l);
--
Gitblit v1.10.0