diff --git a/tools/check_gt_cord.py b/tools/check_gt_cord.py
index 8cd60b4..7b8aa51 100644
--- a/tools/check_gt_cord.py
+++ b/tools/check_gt_cord.py
@@ -6,6 +6,7 @@
 import sys
 import cv2
 import glob
+import shutil
 import pandas as pd
 import numpy as np
 from tqdm import tqdm
@@ -30,6 +31,10 @@ def parse_xml(file):
     box_all = []
     pts = ['xmin', 'ymin', 'xmax', 'ymax']
 
+    # size
+    location = xml.find('location')
+    width = int(location.find('xmax').text) - int(location.find('xmin').text)
+
     # bounding boxes
     for obj in xml.iter('object'):
         bbox = obj.find('bndbox')
@@ -39,7 +44,7 @@ def parse_xml(file):
             cur_pt = int(bbox.find(pt).text) - 1
             bndbox.append(cur_pt)
         box_all += [bndbox]
-    return box_all
+    return box_all, width
 
 def get_box_label(label_df, im_name):
     boxes = []
@@ -58,28 +63,56 @@ def get_box_label(label_df, im_name):
 def _boxvis(img, gt_box_list):
     img1 = img.copy()
     for box in gt_box_list:
-        cv2.rectangle(img1, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 1)
+        cv2.rectangle(img1, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 2)
 
     plt.subplot(1, 1, 1); plt.imshow(img1[:, :, [2,1,0]])
     plt.show()
-    cv2.waitKey()
-    
+    cv2.waitKey(0)
 
-def main():
-    # df = pd.read_csv(src_annotation)
+def _originvis(name, bbox):
+    img = cv2.imread(os.path.join(src_traindir, name))
+    for box in bbox:
+        cv2.rectangle(img, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 1)
+    plt.subplot(1, 1, 1); plt.imshow(img[:, :, [2,1,0]])
+    plt.show()
+    cv2.waitKey(0)
 
-    # for name in df.filename:
-    #     img = cv2.imread(os.path.join(src_traindir, name))
-    #     box, _ = get_box_label(df, name)
-    #     _boxvis(img, box)
+    
+if __name__ == '__main__':
+    vis = False
+
+    df = pd.read_csv(src_annotation)
 
     with open(os.path.join(list_dir, 'train.txt'), 'r') as f:
         img_list = [x.strip() for x in f.readlines()]
 
-    for name in img_list:
-        img = cv2.imread(os.path.join(image_dir, name+'.jpg'))
-        box = parse_xml(os.path.join(anno_dir, name+'.xml'))
-        _boxvis(img, box)
+    filter_list = []
+    for i, name in enumerate(img_list):
+        sys.stdout.write('\rsearch: {:d}/{:d} {:s}'
+                            .format(i + 1, len(img_list), img_list[i]))
+        sys.stdout.flush()
 
-if __name__ == '__main__':
-    main()
+        orgin_name = name.split('_')[0]+'.jpg'
+        orgin_box, _ = get_box_label(df, orgin_name)
+
+        label_w = orgin_box[0][2] - orgin_box[0][0]
+        if label_w > 140 and label_w < 145:
+            if vis:
+                img = cv2.imread(os.path.join(image_dir, name+'.jpg'))
+                box, width = parse_xml(os.path.join(anno_dir, name+'.xml'))
+                _boxvis(img, box)
+                _originvis(orgin_name, orgin_box)
+            else:
+                filter_list.append(orgin_name)
+    
+    if not vis:
+        filter_list = list(set(filter_list))
+        if os.path.exists('temp'):
+            shutil.rmtree('temp')
+        os.mkdir('temp')
+        for name in filter_list:
+            img = cv2.imread(os.path.join(src_traindir, name))
+            orgin_box, _ = get_box_label(df, name)
+            for box in orgin_box:
+                cv2.rectangle(img, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 2)
+            cv2.imwrite('temp/'+name, img)