diff --git a/ppocr/data/imaug/make_shrink_map.py b/ppocr/data/imaug/make_shrink_map.py
index 0e2f3d4f5df..f9899f49e33 100644
--- a/ppocr/data/imaug/make_shrink_map.py
+++ b/ppocr/data/imaug/make_shrink_map.py
@@ -91,6 +91,10 @@ def __call__(self, data):
                     cv2.fillPoly(gt, [shrink.astype(np.int32)], 1)
 
         data["shrink_map"] = gt
+        crop_mask = data["crop_mask"] if "crop_mask" in data else None
+        if crop_mask is not None:
+            # use intersection of crop_mask and src shrink_mask
+            mask = mask * crop_mask
         data["shrink_mask"] = mask
         return data
 
diff --git a/ppocr/data/imaug/random_crop_data.py b/ppocr/data/imaug/random_crop_data.py
index 6625776fa1e..274d84d476e 100644
--- a/ppocr/data/imaug/random_crop_data.py
+++ b/ppocr/data/imaug/random_crop_data.py
@@ -156,11 +156,15 @@ def __call__(self, data):
         scale = min(scale_w, scale_h)
         h = int(crop_h * scale)
         w = int(crop_w * scale)
+        mask = None
         if self.keep_ratio:
             padimg = np.zeros((self.size[1], self.size[0], img.shape[2]), img.dtype)
+            mask = np.zeros((self.size[1], self.size[0]), img.dtype)
             padimg[:h, :w] = cv2.resize(
                 img[crop_y : crop_y + crop_h, crop_x : crop_x + crop_w], (w, h)
             )
+            # mask the padding area
+            mask[:h, :w] = 1
             img = padimg
         else:
             img = cv2.resize(
@@ -181,6 +185,7 @@ def __call__(self, data):
         data["polys"] = np.array(text_polys_crop)
         data["ignore_tags"] = ignore_tags_crop
         data["texts"] = texts_crop
+        data["crop_mask"] = mask
         return data
 
 
diff --git a/ppocr/losses/det_basic_loss.py b/ppocr/losses/det_basic_loss.py
index 5a46e072e25..ea2f854466d 100644
--- a/ppocr/losses/det_basic_loss.py
+++ b/ppocr/losses/det_basic_loss.py
@@ -157,5 +157,8 @@ def __init__(self, reduction="mean"):
         self.reduction = reduction
 
     def forward(self, input, label, mask=None, weight=None, name=None):
+        if mask is not None:
+            input = input * mask
+            label = label * mask
         loss = F.binary_cross_entropy(input, label, reduction=self.reduction)
         return loss