diff --git a/train.py b/train.py
index 0df92b0e..06ebbbe0 100644
--- a/train.py
+++ b/train.py
@@ -12,7 +12,7 @@ from utils.datasets import *
 from utils.utils import *
 
 # Hyperparameters: train.py --evolve --epochs 2 --img-size 320, Metrics: 0.204      0.302      0.175      0.234 (square smart)
-hyp = {'xy': 0.2,  # xy loss gain
+hyp = {'xy': 0.2,  # xy loss gain  (giou is about 0.02)
        'wh': 0.1,  # wh loss gain
        'cls': 0.04,  # cls loss gain
        'conf': 4.5,  # conf loss gain
diff --git a/utils/datasets.py b/utils/datasets.py
index b5f8db6a..98d855e0 100755
--- a/utils/datasets.py
+++ b/utils/datasets.py
@@ -230,7 +230,7 @@ class LoadImagesAndLabels(Dataset):  # for training/testing
         augment_hsv = True
         if self.augment and augment_hsv:
             # SV augmentation by 50%
-            fraction = 0.25  # must be < 1.0
+            fraction = 0.50  # must be < 1.0
             img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)  # hue, sat, val
             S = img_hsv[:, :, 1].astype(np.float32)  # saturation
             V = img_hsv[:, :, 2].astype(np.float32)  # value
@@ -269,7 +269,7 @@ class LoadImagesAndLabels(Dataset):  # for training/testing
 
         # Augment image and labels
         if self.augment:
-            img, labels = random_affine(img, labels, degrees=(-5, 5), translate=(0.0, 0.0), scale=(1.0, 1.0))
+            img, labels = random_affine(img, labels, degrees=(-5, 5), translate=(0.10, 0.10), scale=(0.90, 1.10))
 
         nL = len(labels)  # number of labels
         if nL: