diff --git a/Makefile b/Makefile
index 40bfcecb..cdf200c5 100644
--- a/Makefile
+++ b/Makefile
@@ -1,5 +1,5 @@
-GPU=1
-OPENCV=1
+GPU=0
+OPENCV=0
 DEBUG=0
 
 ARCH= --gpu-architecture=compute_20 --gpu-code=compute_20
@@ -12,7 +12,7 @@ CC=gcc
 NVCC=nvcc
 OPTS=-Ofast
 LDFLAGS= -lm -pthread -lstdc++ 
-COMMON= -I/usr/local/cuda/include/ 
+COMMON= 
 CFLAGS=-Wall -Wfatal-errors 
 
 ifeq ($(DEBUG), 1) 
@@ -29,7 +29,7 @@ COMMON+= `pkg-config --cflags opencv`
 endif
 
 ifeq ($(GPU), 1) 
-COMMON+= -DGPU
+COMMON+= -DGPU -I/usr/local/cuda/include/
 CFLAGS+= -DGPU
 LDFLAGS+= -L/usr/local/cuda/lib64 -lcuda -lcudart -lcublas -lcurand
 endif
diff --git a/cfg/darknet.cfg b/cfg/darknet.cfg
index 64aab1ed..0b3c46c7 100644
--- a/cfg/darknet.cfg
+++ b/cfg/darknet.cfg
@@ -1,5 +1,5 @@
 [net]
-batch=256
+batch=128
 subdivisions=1
 height=256
 width=256
@@ -8,10 +8,10 @@ momentum=0.9
 decay=0.0005
 
 learning_rate=0.01
-policy=step
-scale=.1
-step=100000
-max_batches=400000
+policy=sigmoid
+gamma=.00002
+step=400000
+max_batches=800000
 
 [crop]
 crop_height=224
diff --git a/cfg/yolo.cfg b/cfg/yolo.cfg
index 88176a65..ab46729c 100644
--- a/cfg/yolo.cfg
+++ b/cfg/yolo.cfg
@@ -206,4 +206,5 @@ coords=4
 rescore=0
 joint=0
 objectness=1
+background=0
 
diff --git a/src/box.c b/src/box.c
index 6045d9a7..b99300d3 100644
--- a/src/box.c
+++ b/src/box.c
@@ -2,6 +2,16 @@
 #include <stdio.h>
 #include <math.h>
 
+box float_to_box(float *f)
+{
+    box b;
+    b.x = f[0];
+    b.y = f[1];
+    b.w = f[2];
+    b.h = f[3];
+    return b;
+}
+
 dbox derivative(box a, box b)
 {
     dbox d;
diff --git a/src/box.h b/src/box.h
index f7ef36fd..9b57fb45 100644
--- a/src/box.h
+++ b/src/box.h
@@ -9,6 +9,7 @@ typedef struct{
     float dx, dy, dw, dh;
 } dbox;
 
+box float_to_box(float *f);
 float box_iou(box a, box b);
 float box_rmse(box a, box b);
 dbox diou(box a, box b);
diff --git a/src/compare.c b/src/compare.c
index 0408f800..74c1cf54 100644
--- a/src/compare.c
+++ b/src/compare.c
@@ -241,7 +241,8 @@ void BattleRoyaleWithCheese(char *filename, char *weightfile)
     srand(time(0));
     set_batch_network(&net, 1);
 
-    list *plist = get_paths("data/compare.sort.list");
+    //list *plist = get_paths("data/compare.sort.list");
+    list *plist = get_paths("data/compare.cat.list");
     //list *plist = get_paths("data/compare.val.old");
     char **paths = (char **)list_to_array(plist);
     int N = plist->size;
@@ -256,15 +257,16 @@ void BattleRoyaleWithCheese(char *filename, char *weightfile)
     }
     int round;
     clock_t time=clock();
-    for(round = 1; round <= 40; ++round){
+    for(round = 1; round <= 500; ++round){
         clock_t round_time=clock();
         printf("Round: %d\n", round);
         qsort(boxes, N, sizeof(sortable_bbox), elo_comparator);
         sorta_shuffle(boxes, N, sizeof(sortable_bbox), 10);
+        shuffle(boxes, N, sizeof(sortable_bbox));
         for(i = 0; i < N/2; ++i){
             bbox_fight(boxes+i*2, boxes+i*2+1);
         }
-        if(round >= 4){
+        if(round >= 4 && 0){
             qsort(boxes, N, sizeof(sortable_bbox), elo_comparator);
             if(round == 4){
                 N = N/2;
@@ -275,9 +277,11 @@ void BattleRoyaleWithCheese(char *filename, char *weightfile)
         printf("Round: %f secs, %d remaining\n", sec(clock()-round_time), N);
     }
     qsort(boxes, N, sizeof(sortable_bbox), elo_comparator);
+    FILE *outfp = fopen("results/battle.log", "w");
     for(i = 0; i < N; ++i){
-        printf("%s %f\n", boxes[i].filename, boxes[i].elo);
+        fprintf(outfp, "%s %f\n", boxes[i].filename, boxes[i].elo);
     }
+    fclose(outfp);
     printf("Tournament in %d compares, %f secs\n", total_compares, sec(clock()-time));
 }
 
diff --git a/src/layer.h b/src/layer.h
index 1eb73512..d13cdbfa 100644
--- a/src/layer.h
+++ b/src/layer.h
@@ -60,6 +60,11 @@ typedef struct {
     float beta;
     float kappa;
 
+    float coord_scale;
+    float object_scale;
+    float noobject_scale;
+    float class_scale;
+
     int dontload;
 
     float probability;
diff --git a/src/parser.c b/src/parser.c
index 53e84615..7ea1b3ff 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -182,10 +182,14 @@ region_layer parse_region(list *options, size_params params)
     int num = option_find_int(options, "num", 1);
     int side = option_find_int(options, "side", 7);
     region_layer layer = make_region_layer(params.batch, params.inputs, num, side, classes, coords, rescore);
-    int softmax = option_find_int(options, "softmax", 0);
-    int sqrt = option_find_int(options, "sqrt", 0);
-    layer.softmax = softmax;
-    layer.sqrt = sqrt;
+
+    layer.softmax = option_find_int(options, "softmax", 0);
+    layer.sqrt = option_find_int(options, "sqrt", 0);
+
+    layer.coord_scale = option_find_float(options, "coord_scale", 1);
+    layer.object_scale = option_find_float(options, "object_scale", 1);
+    layer.noobject_scale = option_find_float(options, "noobject_scale", 1);
+    layer.class_scale = option_find_float(options, "class_scale", 1);
     return layer;
 }
 
diff --git a/src/region_layer.c b/src/region_layer.c
index d65c1a87..ecb89c64 100644
--- a/src/region_layer.c
+++ b/src/region_layer.c
@@ -44,15 +44,20 @@ void forward_region_layer(const region_layer l, network_state state)
     int locations = l.side*l.side;
     int i,j;
     memcpy(l.output, state.input, l.outputs*l.batch*sizeof(float));
-    for(i = 0; i < l.batch*locations; ++i){
-        int index = i*((1+l.coords)*l.n + l.classes);
-        if(l.softmax){
-            activate_array(l.output + index, l.n*(1+l.coords), LOGISTIC);
-            int offset = l.n*(1+l.coords);
-            softmax_array(l.output + index + offset, l.classes,
-                    l.output + index + offset);
+    int b;
+    if (l.softmax){
+        for(b = 0; b < l.batch; ++b){
+            int index = b*l.inputs;
+            for (i = 0; i < locations; ++i) {
+                int offset = i*l.classes;
+                softmax_array(l.output + index + offset, l.classes,
+                        l.output + index + offset);
+            }
+            int offset = locations*l.classes;
+            activate_array(l.output + index + offset, locations*l.n*(1+l.coords), LOGISTIC);
         }
     }
+
     if(state.train){
         float avg_iou = 0;
         float avg_cat = 0;
@@ -62,94 +67,91 @@ void forward_region_layer(const region_layer l, network_state state)
         *(l.cost) = 0;
         int size = l.inputs * l.batch;
         memset(l.delta, 0, size * sizeof(float));
-        for (i = 0; i < l.batch*locations; ++i) {
-            int index = i*((1+l.coords)*l.n + l.classes);
-            for(j = 0; j < l.n; ++j){
-                int prob_index = index + j*(1 + l.coords);
-                l.delta[prob_index] = (1./l.n)*(0-l.output[prob_index]);
-                if(l.softmax){
-                    l.delta[prob_index] = 1./(l.n*l.side)*(0-l.output[prob_index]);
-                }
-                *(l.cost) += (1./l.n)*pow(l.output[prob_index], 2);
-                //printf("%f\n", l.output[prob_index]);
-                avg_anyobj += l.output[prob_index];
-            }
-
-            int truth_index = i*(1 + l.coords + l.classes);
-            int best_index = -1;
-            float best_iou = 0;
-            float best_rmse = 4;
-
-            int bg = !state.truth[truth_index];
-            if(bg) {
-                continue;
-            }
-
-            int class_index = index + l.n*(1+l.coords);
-            for(j = 0; j < l.classes; ++j) {
-                l.delta[class_index+j] = state.truth[truth_index+1+j] - l.output[class_index+j];
-                *(l.cost) += pow(state.truth[truth_index+1+j] - l.output[class_index+j], 2);
-                if(state.truth[truth_index + 1 + j]) avg_cat += l.output[class_index+j];
-            }
-            truth_index += l.classes + 1;
-            box truth = {state.truth[truth_index+0], state.truth[truth_index+1], state.truth[truth_index+2], state.truth[truth_index+3]};
-            truth.x /= l.side;
-            truth.y /= l.side;
-
-            for(j = 0; j < l.n; ++j){
-                int out_index = index + j*(1+l.coords);
-                box out = {l.output[out_index+1], l.output[out_index+2], l.output[out_index+3], l.output[out_index+4]};
-
-                out.x /= l.side;
-                out.y /= l.side;
-                if (l.sqrt){
-                    out.w = out.w*out.w;
-                    out.h = out.h*out.h;
+        for (b = 0; b < l.batch; ++b){
+            int index = b*l.inputs;
+            for (i = 0; i < locations; ++i) {
+                int truth_index = (b*locations + i)*(1+l.coords+l.classes);
+                int is_obj = state.truth[truth_index];
+                for (j = 0; j < l.n; ++j) {
+                    int p_index = index + locations*l.classes + i*l.n + j;
+                    l.delta[p_index] = l.noobject_scale*(0 - l.output[p_index]);
+                    *(l.cost) += l.noobject_scale*pow(l.output[p_index], 2);
+                    avg_anyobj += l.output[p_index];
                 }
 
-                float iou  = box_iou(out, truth);
-                float rmse = box_rmse(out, truth);
-                if(best_iou > 0 || iou > 0){
-                    if(iou > best_iou){
-                        best_iou = iou;
-                        best_index = j;
+                int best_index = -1;
+                float best_iou = 0;
+                float best_rmse = 4;
+
+                if (!is_obj) continue;
+
+                int class_index = index + i*l.classes;
+                for(j = 0; j < l.classes; ++j) {
+                    l.delta[class_index+j] = l.class_scale * (state.truth[truth_index+1+j] - l.output[class_index+j]);
+                    *(l.cost) += l.class_scale * pow(state.truth[truth_index+1+j] - l.output[class_index+j], 2);
+                    if(state.truth[truth_index + 1 + j]) avg_cat += l.output[class_index+j];
+                }
+
+                box truth = float_to_box(state.truth + truth_index + 1 + l.classes);
+                truth.x /= l.side;
+                truth.y /= l.side;
+
+                for(j = 0; j < l.n; ++j){
+                    int box_index = index + locations*(l.classes + l.n) + (i*l.n + j) * l.coords;
+                    box out = float_to_box(l.output + box_index);
+                    out.x /= l.side;
+                    out.y /= l.side;
+
+                    if (l.sqrt){
+                        out.w = out.w*out.w;
+                        out.h = out.h*out.h;
                     }
-                }else{
-                    if(rmse < best_rmse){
-                        best_rmse = rmse;
-                        best_index = j;
+
+                    float iou  = box_iou(out, truth);
+                    float rmse = box_rmse(out, truth);
+                    if(best_iou > 0 || iou > 0){
+                        if(iou > best_iou){
+                            best_iou = iou;
+                            best_index = j;
+                        }
+                    }else{
+                        if(rmse < best_rmse){
+                            best_rmse = rmse;
+                            best_index = j;
+                        }
                     }
                 }
-            }
-            //printf("%d", best_index);
-            int in_index = index + best_index*(1+l.coords);
-            *(l.cost) -= pow(l.output[in_index], 2);
-            *(l.cost) += pow(1-l.output[in_index], 2);
-            avg_obj += l.output[in_index];
-            l.delta[in_index+0] = (1.-l.output[in_index]);
-            if(l.softmax){
-                l.delta[in_index+0] = 5*(1.-l.output[in_index]);
-            }
-            //printf("%f\n", l.output[in_index]);
+                int p_index = index + locations*l.classes + i*l.n + best_index;
+                *(l.cost) -= l.noobject_scale * pow(l.output[p_index], 2);
+                *(l.cost) += l.object_scale * pow(1-l.output[p_index], 2);
+                avg_obj += l.output[p_index];
+                l.delta[p_index+0] = l.object_scale * (1.-l.output[p_index]);
 
-            l.delta[in_index+1] = 5*(state.truth[truth_index+0] - l.output[in_index+1]);
-            l.delta[in_index+2] = 5*(state.truth[truth_index+1] - l.output[in_index+2]);
-            if(l.sqrt){
-                l.delta[in_index+3] = 5*(sqrt(state.truth[truth_index+2]) - l.output[in_index+3]);
-                l.delta[in_index+4] = 5*(sqrt(state.truth[truth_index+3]) - l.output[in_index+4]);
-            }else{
-                l.delta[in_index+3] = 5*(state.truth[truth_index+2] - l.output[in_index+3]);
-                l.delta[in_index+4] = 5*(state.truth[truth_index+3] - l.output[in_index+4]);
-            }
+                if(l.rescore){
+                    l.delta[p_index+0] = l.object_scale * (best_iou - l.output[p_index]);
+                }
 
-            *(l.cost) += pow(1-best_iou, 2);
-            avg_iou += best_iou;
-            ++count;
+                int box_index = index + locations*(l.classes + l.n) + (i*l.n + best_index) * l.coords;
+                int tbox_index = truth_index + 1 + l.classes;
+                l.delta[box_index+0] = l.coord_scale*(state.truth[tbox_index + 0] - l.output[box_index + 0]);
+                l.delta[box_index+1] = l.coord_scale*(state.truth[tbox_index + 1] - l.output[box_index + 1]);
+                l.delta[box_index+2] = l.coord_scale*(state.truth[tbox_index + 2] - l.output[box_index + 2]);
+                l.delta[box_index+3] = l.coord_scale*(state.truth[tbox_index + 3] - l.output[box_index + 3]);
+                if(l.sqrt){
+                    l.delta[box_index+2] = l.coord_scale*(sqrt(state.truth[tbox_index + 2]) - l.output[box_index + 2]);
+                    l.delta[box_index+3] = l.coord_scale*(sqrt(state.truth[tbox_index + 3]) - l.output[box_index + 3]);
+                }
+
+                *(l.cost) += pow(1-best_iou, 2);
+                avg_iou += best_iou;
+                ++count;
+            }
             if(l.softmax){
-                gradient_array(l.output + index, l.n*(1+l.coords), LOGISTIC, l.delta + index);
+                gradient_array(l.output + index + locations*l.classes, locations*l.n*(1+l.coords), 
+                        LOGISTIC, l.delta + index + locations*l.classes);
             }
         }
-        printf("Avg IOU: %f, Avg Cat Pred: %f, Avg Obj: %f, Avg Any: %f, count: %d\n", avg_iou/count, avg_cat/count, avg_obj/count, avg_anyobj/(l.batch*locations*l.n), count);
+        printf("Region Avg IOU: %f, Avg Cat Pred: %f, Avg Obj: %f, Avg Any: %f, count: %d\n", avg_iou/count, avg_cat/count, avg_obj/count, avg_anyobj/(l.batch*locations*l.n), count);
     }
 }
 
diff --git a/src/stb_image_write.h b/src/stb_image_write.h
index 10489707..f5250b31 100644
--- a/src/stb_image_write.h
+++ b/src/stb_image_write.h
@@ -279,7 +279,7 @@ void stbiw__write_hdr_scanline(FILE *f, int width, int comp, unsigned char *scra
 {
    unsigned char scanlineheader[4] = { 2, 2, 0, 0 };
    unsigned char rgbe[4];
-   float linear[3];
+   float linear[3] = {0};
    int x;
 
    scanlineheader[2] = (width&0xff00)>>8;
diff --git a/src/swag.c b/src/swag.c
index 4dcf36bf..1398192d 100644
--- a/src/swag.c
+++ b/src/swag.c
@@ -77,7 +77,7 @@ void train_swag(char *cfgfile, char *weightfile)
     int classes = l.classes;
 
     list *plist = get_paths(train_images);
-    int N = plist->size;
+    //int N = plist->size;
     char **paths = (char **)list_to_array(plist);
 
     load_args args = {0};