diff --git a/utils/utils.py b/utils/utils.py
index 262c89c7..6c21bfea 100755
--- a/utils/utils.py
+++ b/utils/utils.py
@@ -369,7 +369,7 @@ def non_max_suppression(prediction, conf_thres=0.5, nms_thres=0.4):
         if prediction.is_cuda:
             unique_labels = unique_labels.cuda(prediction.device)
 
-        nms_style = 'OR'  # 'OR' (default), 'AND', 'MERGE' (experimental)
+        nms_style = 'MERGE'  # 'OR' (default), 'AND', 'MERGE' (experimental)
         for c in unique_labels:
             # Get the detections with class c
             dc = detections[detections[:, -1] == c]
@@ -387,6 +387,12 @@ def non_max_suppression(prediction, conf_thres=0.5, nms_thres=0.4):
                     iou = bbox_iou(det_max[-1], dc[1:])  # iou with other boxes
                     dc = dc[1:][iou < nms_thres]  # remove ious > threshold
 
+                # Image      Total          P          R        mAP
+                #    32       5000      0.633      0.579      0.568
+                #    64       5000      0.619      0.579      0.568
+                #    96       5000      0.652      0.622      0.613
+                #   128       5000      0.651      0.625      0.617
+
             elif nms_style == 'AND':  # requires overlap, single boxes erased
                 while len(dc) > 1:
                     iou = bbox_iou(dc[:1], dc[1:])  # iou with other boxes
@@ -396,10 +402,19 @@ def non_max_suppression(prediction, conf_thres=0.5, nms_thres=0.4):
 
             elif nms_style == 'MERGE':  # weighted mixture box
                 while len(dc) > 0:
-                    if len(dc) == 1:  # Stop if we're at the last detection
-                        det_max.append(dc[:1])  # save highest conf detection
-                        break
-                    iou = bbox_iou(dc[:1], dc[1:])  # iou with other boxes
+                    iou = bbox_iou(dc[:1], dc[0:])  # iou with other boxes
+                    i = iou > nms_thres
+
+                    weights = dc[i, 4:5] * dc[i, 5:6]
+                    dc[0, :4] = (weights * dc[i, :4]).sum(0) / weights.sum()
+                    det_max.append(dc[:1])
+                    dc = dc[iou < nms_thres]
+
+                # Image      Total          P          R        mAP
+                #    32       5000      0.635      0.581      0.569
+                #    64       5000       0.63      0.591      0.578
+                #    96       5000       0.66       0.63       0.62
+                #   128       5000      0.657      0.631      0.622
 
             if len(det_max) > 0:
                 det_max = torch.cat(det_max)