diff --git a/src/data.c b/src/data.c
index 8dd7d9a6309f51c7fcb36ad4af177273dc6083db..70692487b72f8565257a7b1f2eecd55eddeb8699 100644
--- a/src/data.c
+++ b/src/data.c
@@ -137,18 +137,20 @@ void fill_truth_detection(char *path, float *truth, int classes, int height, int
if(j < 0) j = 0;
if(j >= num_height) j = num_height-1;
- float dw = (x - i*box_width)/box_width;
- float dh = (y - j*box_height)/box_height;
+ float dw = constrain(0,1, (x - i*box_width)/box_width );
+ float dh = constrain(0,1, (y - j*box_height)/box_height );
+ float th = constrain(0,1, h*(height+jitter)/height );
+ float tw = constrain(0,1, w*(width+jitter)/width );
int index = (i+j*num_width)*(4+classes+background);
- if(truth[index+classes+background]) continue;
+ if(truth[index+classes+background+2]) continue;
if(background) truth[index++] = 0;
truth[index+id] = 1;
index += classes;
truth[index++] = dh;
truth[index++] = dw;
- truth[index++] = h*(height+jitter)/height;
- truth[index++] = w*(width+jitter)/width;
+ truth[index++] = th;
+ truth[index++] = tw;
}
free(boxes);
}
diff --git a/src/detection.c b/src/detection.c
index 522a3219a51aed544a0ff4337f27c817c7fc0d73..1800ca6ca88b05188d790c7456dd029b39a89861 100644
--- a/src/detection.c
+++ b/src/detection.c
@@ -50,7 +50,7 @@ void train_detection(char *cfgfile, char *weightfile)
if(weightfile){
load_weights(&net, weightfile);
}
- net.seen = 0;
+ //net.seen = 0;
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = 128;
srand(time(0));
@@ -63,7 +63,7 @@ void train_detection(char *cfgfile, char *weightfile)
int im_dim = 512;
int jitter = 64;
int classes = 20;
- int background = 0;
+ int background = 1;
pthread_t load_thread = load_data_detection_thread(imgs, paths, plist->size, classes, im_dim, im_dim, 7, 7, jitter, background, &buffer);
clock_t time;
while(1){
@@ -109,8 +109,9 @@ void validate_detection(char *cfgfile, char *weightfile)
char **paths = (char **)list_to_array(plist);
int im_size = 448;
int classes = 20;
- int background = 0;
- int num_output = 7*7*(4+classes+background);
+ int background = 1;
+ int nuisance = 0;
+ int num_output = 7*7*(4+classes+background+nuisance);
int m = plist->size;
int i = 0;
@@ -134,17 +135,19 @@ void validate_detection(char *cfgfile, char *weightfile)
matrix pred = network_predict_data(net, val);
int j, k, class;
for(j = 0; j < pred.rows; ++j){
- for(k = 0; k < pred.cols; k += classes+4+background){
+ for(k = 0; k < pred.cols; k += classes+4+background+nuisance){
+ float scale = 1.;
+ if(nuisance) scale = pred.vals[j][k];
for(class = 0; class < classes; ++class){
- int index = (k)/(classes+4+background);
+ int index = (k)/(classes+4+background+nuisance);
int r = index/7;
int c = index%7;
- int ci = k+classes+background;
+ int ci = k+classes+background+nuisance;
float y = (r + pred.vals[j][ci + 0])/7.;
float x = (c + pred.vals[j][ci + 1])/7.;
float h = pred.vals[j][ci + 2];
float w = pred.vals[j][ci + 3];
- printf("%d %d %f %f %f %f %f\n", (i-1)*m/splits + j, class, pred.vals[j][k+class+background], y, x, h, w);
+ printf("%d %d %f %f %f %f %f\n", (i-1)*m/splits + j, class, scale*pred.vals[j][k+class+background+nuisance], y, x, h, w);
}
}
}
diff --git a/src/detection_layer.c b/src/detection_layer.c
index 0a754fd4b3bef077606efd184d4ad56143ad832b..27a4daf345221d1033f10509213759c0f7ea3347 100644
--- a/src/detection_layer.c
+++ b/src/detection_layer.c
@@ -16,7 +16,7 @@ int get_detection_layer_output_size(detection_layer layer)
return get_detection_layer_locations(layer)*(layer.background + layer.classes + layer.coords);
}
-detection_layer *make_detection_layer(int batch, int inputs, int classes, int coords, int rescore, int background)
+detection_layer *make_detection_layer(int batch, int inputs, int classes, int coords, int rescore, int background, int nuisance)
{
detection_layer *layer = calloc(1, sizeof(detection_layer));
@@ -25,6 +25,7 @@ detection_layer *make_detection_layer(int batch, int inputs, int classes, int co
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));
@@ -72,12 +73,18 @@ void forward_detection_layer(const detection_layer layer, network_state state)
int mask = (!state.truth || state.truth[out_i + layer.background + layer.classes + 2]);
float scale = 1;
if(layer.rescore) scale = state.input[in_i++];
- if(layer.background) layer.output[out_i++] = 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.background){
+ 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);
}
@@ -85,6 +92,7 @@ void forward_detection_layer(const detection_layer layer, network_state state)
layer.output[out_i++] = mask*state.input[in_i++];
}
}
+ /*
if(layer.background || 1){
for(i = 0; i < layer.batch*locations; ++i){
int index = i*(layer.classes+layer.coords+layer.background);
@@ -95,6 +103,7 @@ void forward_detection_layer(const detection_layer layer, network_state state)
}
}
}
+ */
}
void backward_detection_layer(const detection_layer layer, network_state state)
@@ -107,13 +116,15 @@ void backward_detection_layer(const detection_layer layer, network_state state)
float scale = 1;
float latent_delta = 0;
if(layer.rescore) scale = state.input[in_i++];
- if(layer.background) state.delta[in_i++] = scale*layer.delta[out_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.background) gradient_array(layer.output + out_i, layer.coords, LOGISTIC, 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++];
}
diff --git a/src/detection_layer.h b/src/detection_layer.h
index 2ad1ef2350528485586f327839c3da5483a3b2b4..7be576e80fa15dcd45c433900828fb6fabb82221 100644
--- a/src/detection_layer.h
+++ b/src/detection_layer.h
@@ -10,6 +10,7 @@ typedef struct {
int coords;
int background;
int rescore;
+ int nuisance;
float *output;
float *delta;
#ifdef GPU
@@ -18,7 +19,7 @@ typedef struct {
#endif
} detection_layer;
-detection_layer *make_detection_layer(int batch, int inputs, int classes, int coords, int rescore, int background);
+detection_layer *make_detection_layer(int batch, int inputs, int classes, int coords, int rescore, int background, int nuisance);
void forward_detection_layer(const detection_layer layer, network_state state);
void backward_detection_layer(const detection_layer layer, network_state state);
int get_detection_layer_output_size(detection_layer layer);
diff --git a/src/parser.c b/src/parser.c
index 6ff978c7c3a8199549266bb3712081bcf6d0318f..e4ee17e4ddefa608001de325ce146c4f8745f2a8 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -165,8 +165,9 @@ detection_layer *parse_detection(list *options, size_params params)
int coords = option_find_int(options, "coords", 1);
int classes = option_find_int(options, "classes", 1);
int rescore = option_find_int(options, "rescore", 1);
+ int nuisance = option_find_int(options, "nuisance", 0);
int background = option_find_int(options, "background", 1);
- detection_layer *layer = make_detection_layer(params.batch, params.inputs, classes, coords, rescore, background);
+ detection_layer *layer = make_detection_layer(params.batch, params.inputs, classes, coords, rescore, background, nuisance);
option_unused(options);
return layer;
}
@@ -550,7 +551,7 @@ void print_softmax_cfg(FILE *fp, softmax_layer *l, network net, int count)
void print_detection_cfg(FILE *fp, detection_layer *l, network net, int count)
{
fprintf(fp, "[detection]\n");
- fprintf(fp, "classes=%d\ncoords=%d\nrescore=%d\n", l->classes, l->coords, l->rescore);
+ fprintf(fp, "classes=%d\ncoords=%d\nrescore=%d\nnuisance=%d\n", l->classes, l->coords, l->rescore, l->nuisance);
fprintf(fp, "\n");
}
diff --git a/src/utils.c b/src/utils.c
index 6fb0e439815b930db61dc93332a982f601cedaa2..bc39ceccb7af11737491217ecff14593346fdc93 100644
--- a/src/utils.c
+++ b/src/utils.c
@@ -276,10 +276,10 @@ float variance_array(float *a, int n)
return variance;
}
-float constrain(float a, float max)
+float constrain(float min, float max, float a)
{
- if(a > abs(max)) return abs(max);
- if(a < -abs(max)) return -abs(max);
+ if (a < min) return min;
+ if (a > max) return max;
return a;
}
diff --git a/src/utils.h b/src/utils.h
index 4c6b2a9faee725a6506b640b67f0cbccc9728b6b..578abc37ee3fd2e76707294158bb227cf6bcf4f9 100644
--- a/src/utils.h
+++ b/src/utils.h
@@ -26,7 +26,7 @@ void normalize_array(float *a, int n);
void scale_array(float *a, int n, float s);
void translate_array(float *a, int n, float s);
int max_index(float *a, int n);
-float constrain(float a, float max);
+float constrain(float min, float max, float a);
float mse_array(float *a, int n);
float rand_normal();
float rand_uniform();