From 4f50e29365c8b8fd3aa9b67167701c1ada1e373f Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Thu, 09 Apr 2015 22:18:54 +0000
Subject: [PATCH] big change to images
---
src/detection_layer.c | 221 +++++++++++++++++++++++++++++++++++++++++++------------
1 files changed, 173 insertions(+), 48 deletions(-)
diff --git a/src/detection_layer.c b/src/detection_layer.c
index 6537079..73b2862 100644
--- a/src/detection_layer.c
+++ b/src/detection_layer.c
@@ -1,72 +1,197 @@
-int detection_out_height(detection_layer layer)
+#include "detection_layer.h"
+#include "activations.h"
+#include "softmax_layer.h"
+#include "blas.h"
+#include "cuda.h"
+#include <stdio.h>
+#include <stdlib.h>
+
+int get_detection_layer_locations(detection_layer layer)
{
- return layer.size + layer.h*layer.stride;
+ return layer.inputs / (layer.classes+layer.coords+layer.rescore+layer.background);
}
-int detection_out_width(detection_layer layer)
+int get_detection_layer_output_size(detection_layer layer)
{
- return layer.size + layer.w*layer.stride;
+ return get_detection_layer_locations(layer)*(layer.background + layer.classes + layer.coords);
}
-detection_layer *make_detection_layer(int batch, int h, int w, int c, int n, int size, int stride, ACTIVATION activation)
+detection_layer *make_detection_layer(int batch, int inputs, int classes, int coords, int rescore, int background, int nuisance)
{
- int i;
- size = 2*(size/2)+1; //HA! And you thought you'd use an even sized filter...
detection_layer *layer = calloc(1, sizeof(detection_layer));
- layer->h = h;
- layer->w = w;
- layer->c = c;
- layer->n = n;
+
layer->batch = batch;
- layer->stride = stride;
- layer->size = size;
- assert(c%n == 0);
+ layer->inputs = inputs;
+ layer->classes = classes;
+ layer->coords = coords;
+ layer->rescore = rescore;
+ layer->nuisance = nuisance;
+ layer->background = background;
+ int outputs = get_detection_layer_output_size(*layer);
+ layer->output = calloc(batch*outputs, sizeof(float));
+ layer->delta = calloc(batch*outputs, sizeof(float));
+ #ifdef GPU
+ layer->output_gpu = cuda_make_array(0, batch*outputs);
+ layer->delta_gpu = cuda_make_array(0, batch*outputs);
+ #endif
- layer->filters = calloc(c*size*size, sizeof(float));
- layer->filter_updates = calloc(c*size*size, sizeof(float));
- layer->filter_momentum = calloc(c*size*size, sizeof(float));
-
- float scale = 1./(size*size*c);
- for(i = 0; i < c*n*size*size; ++i) layer->filters[i] = scale*(rand_uniform());
-
- int out_h = detection_out_height(*layer);
- int out_w = detection_out_width(*layer);
-
- layer->output = calloc(layer->batch * out_h * out_w * n, sizeof(float));
- layer->delta = calloc(layer->batch * out_h * out_w * n, sizeof(float));
-
- layer->activation = activation;
-
- fprintf(stderr, "Convolutional Layer: %d x %d x %d image, %d filters -> %d x %d x %d image\n", h,w,c,n, out_h, out_w, n);
+ fprintf(stderr, "Detection Layer\n");
srand(0);
return layer;
}
-void forward_detection_layer(const detection_layer layer, float *in)
+void dark_zone(detection_layer layer, int class, int start, network_state state)
{
- int out_h = detection_out_height(layer);
- int out_w = detection_out_width(layer);
- int i,j,fh, fw,c;
- memset(layer.output, 0, layer->batch*layer->n*out_h*out_w*sizeof(float));
- for(c = 0; c < layer.c; ++c){
- for(i = 0; i < layer.h; ++i){
- for(j = 0; j < layer.w; ++j){
- float val = layer->input[j+(i + c*layer.h)*layer.w];
- for(fh = 0; fh < layer.size; ++fh){
- for(fw = 0; fw < layer.size; ++fw){
- int h = i*layer.stride + fh;
- int w = j*layer.stride + fw;
- layer.output[w+(h+c/n*out_h)*out_w] += val*layer->filters[fw+(fh+c*layer.size)*layer.size];
- }
- }
- }
+ int index = start+layer.background+class;
+ int size = layer.classes+layer.coords+layer.background;
+ int location = (index%(7*7*size)) / size ;
+ int r = location / 7;
+ int c = location % 7;
+ int dr, dc;
+ for(dr = -1; dr <= 1; ++dr){
+ for(dc = -1; dc <= 1; ++dc){
+ if(!(dr || dc)) continue;
+ if((r + dr) > 6 || (r + dr) < 0) continue;
+ if((c + dc) > 6 || (c + dc) < 0) continue;
+ int di = (dr*7 + dc) * size;
+ if(state.truth[index+di]) continue;
+ layer.output[index + di] = 0;
+ //if(!state.truth[start+di]) continue;
+ //layer.output[start + di] = 1;
}
}
}
-void backward_detection_layer(const detection_layer layer, float *delta)
+void forward_detection_layer(const detection_layer layer, network_state state)
{
+ int in_i = 0;
+ int out_i = 0;
+ int locations = get_detection_layer_locations(layer);
+ int i,j;
+ for(i = 0; i < layer.batch*locations; ++i){
+ int mask = (!state.truth || state.truth[out_i + layer.background + layer.classes + 2]);
+ float scale = 1;
+ if(layer.rescore) scale = state.input[in_i++];
+ else if(layer.nuisance){
+ layer.output[out_i++] = 1-state.input[in_i++];
+ scale = mask;
+ }
+ else if(layer.background) layer.output[out_i++] = scale*state.input[in_i++];
+
+ for(j = 0; j < layer.classes; ++j){
+ layer.output[out_i++] = scale*state.input[in_i++];
+ }
+ if(layer.nuisance){
+
+ }else if(layer.background){
+ softmax_array(layer.output + out_i - layer.classes-layer.background, layer.classes+layer.background, layer.output + out_i - layer.classes-layer.background);
+ activate_array(state.input+in_i, layer.coords, LOGISTIC);
+ }
+ for(j = 0; j < layer.coords; ++j){
+ layer.output[out_i++] = mask*state.input[in_i++];
+ }
+ }
+ /*
+ int count = 0;
+ for(i = 0; i < layer.batch*locations; ++i){
+ for(j = 0; j < layer.classes+layer.background; ++j){
+ printf("%f, ", layer.output[count++]);
+ }
+ printf("\n");
+ for(j = 0; j < layer.coords; ++j){
+ printf("%f, ", layer.output[count++]);
+ }
+ printf("\n");
+ }
+ */
+ /*
+ if(layer.background || 1){
+ for(i = 0; i < layer.batch*locations; ++i){
+ int index = i*(layer.classes+layer.coords+layer.background);
+ for(j= 0; j < layer.classes; ++j){
+ if(state.truth[index+j+layer.background]){
+ //dark_zone(layer, j, index, state);
+ }
+ }
+ }
+ }
+ */
}
+void backward_detection_layer(const detection_layer layer, network_state state)
+{
+ int locations = get_detection_layer_locations(layer);
+ int i,j;
+ int in_i = 0;
+ int out_i = 0;
+ for(i = 0; i < layer.batch*locations; ++i){
+ float scale = 1;
+ float latent_delta = 0;
+ if(layer.rescore) scale = state.input[in_i++];
+ else if (layer.nuisance) state.delta[in_i++] = -layer.delta[out_i++];
+ else if (layer.background) state.delta[in_i++] = scale*layer.delta[out_i++];
+ for(j = 0; j < layer.classes; ++j){
+ latent_delta += state.input[in_i]*layer.delta[out_i];
+ state.delta[in_i++] = scale*layer.delta[out_i++];
+ }
+
+ if (layer.nuisance) {
+
+ }else if (layer.background) gradient_array(layer.output + out_i, layer.coords, LOGISTIC, layer.delta + out_i);
+ for(j = 0; j < layer.coords; ++j){
+ state.delta[in_i++] = layer.delta[out_i++];
+ }
+ if(layer.rescore) state.delta[in_i-layer.coords-layer.classes-layer.rescore-layer.background] = latent_delta;
+ }
+}
+
+#ifdef GPU
+
+void forward_detection_layer_gpu(const detection_layer layer, network_state state)
+{
+ int outputs = get_detection_layer_output_size(layer);
+ float *in_cpu = calloc(layer.batch*layer.inputs, sizeof(float));
+ float *truth_cpu = 0;
+ if(state.truth){
+ truth_cpu = calloc(layer.batch*outputs, sizeof(float));
+ cuda_pull_array(state.truth, truth_cpu, layer.batch*outputs);
+ }
+ cuda_pull_array(state.input, in_cpu, layer.batch*layer.inputs);
+ network_state cpu_state;
+ cpu_state.train = state.train;
+ cpu_state.truth = truth_cpu;
+ cpu_state.input = in_cpu;
+ forward_detection_layer(layer, cpu_state);
+ cuda_push_array(layer.output_gpu, layer.output, layer.batch*outputs);
+ free(cpu_state.input);
+ if(cpu_state.truth) free(cpu_state.truth);
+}
+
+void backward_detection_layer_gpu(detection_layer layer, network_state state)
+{
+ int outputs = get_detection_layer_output_size(layer);
+
+ float *in_cpu = calloc(layer.batch*layer.inputs, sizeof(float));
+ float *delta_cpu = calloc(layer.batch*layer.inputs, sizeof(float));
+ float *truth_cpu = 0;
+ if(state.truth){
+ truth_cpu = calloc(layer.batch*outputs, sizeof(float));
+ cuda_pull_array(state.truth, truth_cpu, layer.batch*outputs);
+ }
+ network_state cpu_state;
+ cpu_state.train = state.train;
+ cpu_state.input = in_cpu;
+ cpu_state.truth = truth_cpu;
+ cpu_state.delta = delta_cpu;
+
+ cuda_pull_array(state.input, in_cpu, layer.batch*layer.inputs);
+ cuda_pull_array(layer.delta_gpu, layer.delta, layer.batch*outputs);
+ backward_detection_layer(layer, cpu_state);
+ cuda_push_array(state.delta, delta_cpu, layer.batch*layer.inputs);
+
+ free(in_cpu);
+ free(delta_cpu);
+}
+#endif
--
Gitblit v1.10.0