From 9a01e6ccb7a74ff77e99060cf18acd6cfdb74b8e Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Fri, 11 Nov 2016 16:48:40 +0000
Subject: [PATCH] :fire: crush. crush. admit. :fire:
---
src/region_layer.c | 213 ++++++++++++++++++++++++++++++++++-------------------
1 files changed, 136 insertions(+), 77 deletions(-)
diff --git a/src/region_layer.c b/src/region_layer.c
index 5f8b3cc..ac30e88 100644
--- a/src/region_layer.c
+++ b/src/region_layer.c
@@ -48,13 +48,18 @@
return l;
}
+#define DOABS 1
box get_region_box(float *x, float *biases, int n, int index, int i, int j, int w, int h)
{
box b;
- b.x = (i + .5)/w + x[index + 0] * biases[2*n];
- b.y = (j + .5)/h + x[index + 1] * biases[2*n + 1];
+ b.x = (i + logistic_activate(x[index + 0])) / w;
+ b.y = (j + logistic_activate(x[index + 1])) / h;
b.w = exp(x[index + 2]) * biases[2*n];
b.h = exp(x[index + 3]) * biases[2*n+1];
+ if(DOABS){
+ b.w = exp(x[index + 2]) * biases[2*n] / w;
+ b.h = exp(x[index + 3]) * biases[2*n+1] / h;
+ }
return b;
}
@@ -63,18 +68,47 @@
box pred = get_region_box(x, biases, n, index, i, j, w, h);
float iou = box_iou(pred, truth);
- float tx = (truth.x - (i + .5)/w) / biases[2*n];
- float ty = (truth.y - (j + .5)/h) / biases[2*n + 1];
+ float tx = (truth.x*w - i);
+ float ty = (truth.y*h - j);
float tw = log(truth.w / biases[2*n]);
float th = log(truth.h / biases[2*n + 1]);
+ if(DOABS){
+ tw = log(truth.w*w / biases[2*n]);
+ th = log(truth.h*h / biases[2*n + 1]);
+ }
- delta[index + 0] = scale * (tx - x[index + 0]);
- delta[index + 1] = scale * (ty - x[index + 1]);
+ delta[index + 0] = scale * (tx - logistic_activate(x[index + 0])) * logistic_gradient(logistic_activate(x[index + 0]));
+ delta[index + 1] = scale * (ty - logistic_activate(x[index + 1])) * logistic_gradient(logistic_activate(x[index + 1]));
delta[index + 2] = scale * (tw - x[index + 2]);
delta[index + 3] = scale * (th - x[index + 3]);
return iou;
}
+void delta_region_class(float *output, float *delta, int index, int class, int classes, tree *hier, float scale, float *avg_cat)
+{
+ int i, n;
+ if(hier){
+ float pred = 1;
+ while(class >= 0){
+ pred *= output[index + class];
+ int g = hier->group[class];
+ int offset = hier->group_offset[g];
+ for(i = 0; i < hier->group_size[g]; ++i){
+ delta[index + offset + i] = scale * (0 - output[index + offset + i]);
+ }
+ delta[index + class] = scale * (1 - output[index + class]);
+
+ class = hier->parent[class];
+ }
+ *avg_cat += pred;
+ } else {
+ for(n = 0; n < classes; ++n){
+ delta[index + n] = scale * (((n == class)?1 : 0) - output[index + n]);
+ if(n == class) *avg_cat += output[index + n];
+ }
+ }
+}
+
float logit(float x)
{
return log(x/(1.-x));
@@ -85,8 +119,7 @@
return (x != x);
}
-#define LOG 0
-
+void softmax_tree(float *input, int batch, int inputs, float temp, tree *hierarchy, float *output);
void forward_region_layer(const region_layer l, network_state state)
{
int i,j,b,t,n;
@@ -97,7 +130,33 @@
for(i = 0; i < l.h*l.w*l.n; ++i){
int index = size*i + b*l.outputs;
l.output[index + 4] = logistic_activate(l.output[index + 4]);
- if(l.softmax){
+ }
+ }
+
+
+ if (l.softmax_tree){
+#ifdef GPU
+ cuda_push_array(l.output_gpu, l.output, l.batch*l.outputs);
+ int i;
+ int count = 5;
+ for (i = 0; i < l.softmax_tree->groups; ++i) {
+ int group_size = l.softmax_tree->group_size[i];
+ softmax_gpu(l.output_gpu+count, group_size, l.classes + 5, l.w*l.h*l.n*l.batch, 1, l.output_gpu + count);
+ count += group_size;
+ }
+ cuda_pull_array(l.output_gpu, l.output, l.batch*l.outputs);
+#else
+ for (b = 0; b < l.batch; ++b){
+ for(i = 0; i < l.h*l.w*l.n; ++i){
+ int index = size*i + b*l.outputs;
+ softmax_tree(l.output + index + 5, 1, 0, 1, l.softmax_tree, l.output + index + 5);
+ }
+ }
+#endif
+ } else if (l.softmax){
+ for (b = 0; b < l.batch; ++b){
+ for(i = 0; i < l.h*l.w*l.n; ++i){
+ int index = size*i + b*l.outputs;
softmax(l.output + index + 5, l.classes, 1, l.output + index + 5);
}
}
@@ -110,6 +169,7 @@
float avg_obj = 0;
float avg_anyobj = 0;
int count = 0;
+ int class_count = 0;
*(l.cost) = 0;
for (b = 0; b < l.batch; ++b) {
for (j = 0; j < l.h; ++j) {
@@ -118,34 +178,47 @@
int index = size*(j*l.w*l.n + i*l.n + n) + b*l.outputs;
box pred = get_region_box(l.output, l.biases, n, index, i, j, l.w, l.h);
float best_iou = 0;
+ int best_class = -1;
for(t = 0; t < 30; ++t){
box truth = float_to_box(state.truth + t*5 + b*l.truths);
if(!truth.x) break;
float iou = box_iou(pred, truth);
- if (iou > best_iou) best_iou = iou;
+ if (iou > best_iou) {
+ best_class = state.truth[t*5 + b*l.truths + 4];
+ best_iou = iou;
+ }
}
avg_anyobj += l.output[index + 4];
l.delta[index + 4] = l.noobject_scale * ((0 - l.output[index + 4]) * logistic_gradient(l.output[index + 4]));
- if(best_iou > .5) l.delta[index + 4] = 0;
+ if(l.classfix == -1) l.delta[index + 4] = l.noobject_scale * ((best_iou - l.output[index + 4]) * logistic_gradient(l.output[index + 4]));
+ else{
+ if (best_iou > l.thresh) {
+ l.delta[index + 4] = 0;
+ if(l.classfix > 0){
+ delta_region_class(l.output, l.delta, index + 5, best_class, l.classes, l.softmax_tree, l.class_scale*(l.classfix == 2 ? l.output[index + 4] : 1), &avg_cat);
+ ++class_count;
+ }
+ }
+ }
- if(*(state.net.seen) < 6400){
+ if(*(state.net.seen) < 12800){
box truth = {0};
truth.x = (i + .5)/l.w;
truth.y = (j + .5)/l.h;
- truth.w = .5;
- truth.h = .5;
+ truth.w = l.biases[2*n];
+ truth.h = l.biases[2*n+1];
+ if(DOABS){
+ truth.w = l.biases[2*n]/l.w;
+ truth.h = l.biases[2*n+1]/l.h;
+ }
delta_region_box(truth, l.output, l.biases, n, index, i, j, l.w, l.h, l.delta, .01);
- //l.delta[index + 0] = .1 * (0 - l.output[index + 0]);
- //l.delta[index + 1] = .1 * (0 - l.output[index + 1]);
- //l.delta[index + 2] = .1 * (0 - l.output[index + 2]);
- //l.delta[index + 3] = .1 * (0 - l.output[index + 3]);
}
}
}
}
for(t = 0; t < 30; ++t){
box truth = float_to_box(state.truth + t*5 + b*l.truths);
- int class = state.truth[t*5 + b*l.truths + 4];
+
if(!truth.x) break;
float best_iou = 0;
int best_index = 0;
@@ -156,11 +229,19 @@
box truth_shift = truth;
truth_shift.x = 0;
truth_shift.y = 0;
- printf("index %d %d\n",i, j);
+ //printf("index %d %d\n",i, j);
for(n = 0; n < l.n; ++n){
int index = size*(j*l.w*l.n + i*l.n + n) + b*l.outputs;
box pred = get_region_box(l.output, l.biases, n, index, i, j, l.w, l.h);
- printf("pred: (%f, %f) %f x %f\n", pred.x*l.w - i - .5, pred.y * l.h - j - .5, pred.w, pred.h);
+ if(l.bias_match){
+ pred.w = l.biases[2*n];
+ pred.h = l.biases[2*n+1];
+ if(DOABS){
+ pred.w = l.biases[2*n]/l.w;
+ pred.h = l.biases[2*n+1]/l.h;
+ }
+ }
+ //printf("pred: (%f, %f) %f x %f\n", pred.x, pred.y, pred.w, pred.h);
pred.x = 0;
pred.y = 0;
float iou = box_iou(pred, truth_shift);
@@ -170,7 +251,7 @@
best_n = n;
}
}
- printf("%d %f (%f, %f) %f x %f\n", best_n, best_iou, truth.x * l.w - i - .5, truth.y*l.h - j - .5, truth.w, truth.h);
+ //printf("%d %f (%f, %f) %f x %f\n", best_n, best_iou, truth.x, truth.y, truth.w, truth.h);
float iou = delta_region_box(truth, l.output, l.biases, best_n, best_index, i, j, l.w, l.h, l.delta, l.coord_scale);
if(iou > .5) recall += 1;
@@ -182,48 +263,19 @@
if (l.rescore) {
l.delta[best_index + 4] = l.object_scale * (iou - l.output[best_index + 4]) * logistic_gradient(l.output[best_index + 4]);
}
- //printf("%f\n", l.delta[best_index+1]);
- /*
- if(isnan(l.delta[best_index+1])){
- printf("%f\n", true_scale);
- printf("%f\n", l.output[best_index + 1]);
- printf("%f\n", truth.w);
- printf("%f\n", truth.h);
- error("bad");
- }
- */
- for(n = 0; n < l.classes; ++n){
- l.delta[best_index + 5 + n] = l.class_scale * (((n == class)?1 : 0) - l.output[best_index + 5 + n]);
- if(n == class) avg_cat += l.output[best_index + 5 + n];
- }
- /*
- if(0){
- printf("truth: %f %f %f %f\n", truth.x, truth.y, truth.w, truth.h);
- printf("pred: %f %f %f %f\n\n", pred.x, pred.y, pred.w, pred.h);
- float aspect = exp(true_aspect);
- float scale = logistic_activate(true_scale);
- float move_x = true_dx;
- float move_y = true_dy;
- box b;
- b.w = sqrt(scale * aspect);
- b.h = b.w * 1./aspect;
- b.x = move_x * b.w + (i + .5)/l.w;
- b.y = move_y * b.h + (j + .5)/l.h;
- printf("%f %f\n", b.x, truth.x);
- printf("%f %f\n", b.y, truth.y);
- printf("%f %f\n", b.w, truth.w);
- printf("%f %f\n", b.h, truth.h);
- //printf("%f\n", box_iou(b, truth));
- }
- */
+
+ int class = state.truth[t*5 + b*l.truths + 4];
+ if (l.map) class = l.map[class];
+ delta_region_class(l.output, l.delta, best_index + 5, class, l.classes, l.softmax_tree, l.class_scale, &avg_cat);
++count;
+ ++class_count;
}
}
- printf("\n");
+ //printf("\n");
reorg(l.delta, l.w*l.h, size*l.n, l.batch, 0);
*(l.cost) = pow(mag_array(l.delta, l.outputs * l.batch), 2);
- printf("Region Avg IOU: %f, Class: %f, Obj: %f, No Obj: %f, Avg Recall: %f, count: %d\n", avg_iou/count, avg_cat/count, avg_obj/count, avg_anyobj/(l.w*l.h*l.n*l.batch), recall/count, count);
+ printf("Region Avg IOU: %f, Class: %f, Obj: %f, No Obj: %f, Avg Recall: %f, count: %d\n", avg_iou/count, avg_cat/class_count, avg_obj/count, avg_anyobj/(l.w*l.h*l.n*l.batch), recall/count, count);
}
void backward_region_layer(const region_layer l, network_state state)
@@ -235,7 +287,6 @@
{
int i,j,n;
float *predictions = l.output;
- //int per_cell = 5*num+classes;
for (i = 0; i < l.w*l.h; ++i){
int row = i / l.w;
int col = i % l.w;
@@ -243,25 +294,33 @@
int index = i*l.n + n;
int p_index = index * (l.classes + 5) + 4;
float scale = predictions[p_index];
+ if(l.classfix == -1 && scale < .5) scale = 0;
int box_index = index * (l.classes + 5);
- boxes[index].x = (predictions[box_index + 0] + col + .5) / l.w * w;
- boxes[index].y = (predictions[box_index + 1] + row + .5) / l.h * h;
- if(0){
- boxes[index].x = (logistic_activate(predictions[box_index + 0]) + col) / l.w * w;
- boxes[index].y = (logistic_activate(predictions[box_index + 1]) + row) / l.h * h;
- }
- boxes[index].w = pow(logistic_activate(predictions[box_index + 2]), (l.sqrt?2:1)) * w;
- boxes[index].h = pow(logistic_activate(predictions[box_index + 3]), (l.sqrt?2:1)) * h;
- if(1){
- boxes[index].x = ((col + .5)/l.w + predictions[box_index + 0] * .5) * w;
- boxes[index].y = ((row + .5)/l.h + predictions[box_index + 1] * .5) * h;
- boxes[index].w = (exp(predictions[box_index + 2]) * .5) * w;
- boxes[index].h = (exp(predictions[box_index + 3]) * .5) * h;
- }
- for(j = 0; j < l.classes; ++j){
- int class_index = index * (l.classes + 5) + 5;
- float prob = scale*predictions[class_index+j];
- probs[index][j] = (prob > thresh) ? prob : 0;
+ boxes[index] = get_region_box(predictions, l.biases, n, box_index, col, row, l.w, l.h);
+ boxes[index].x *= w;
+ boxes[index].y *= h;
+ boxes[index].w *= w;
+ boxes[index].h *= h;
+
+ int class_index = index * (l.classes + 5) + 5;
+ if(l.softmax_tree){
+
+ hierarchy_predictions(predictions + class_index, l.classes, l.softmax_tree, 0);
+ int found = 0;
+ for(j = l.classes - 1; j >= 0; --j){
+ if(!found && predictions[class_index + j] > .5){
+ found = 1;
+ } else {
+ predictions[class_index + j] = 0;
+ }
+ float prob = predictions[class_index+j];
+ probs[index][j] = (scale > thresh) ? prob : 0;
+ }
+ }else{
+ for(j = 0; j < l.classes; ++j){
+ float prob = scale*predictions[class_index+j];
+ probs[index][j] = (prob > thresh) ? prob : 0;
+ }
}
if(only_objectness){
probs[index][0] = scale;
--
Gitblit v1.10.0