Update app.py

app.py

@@ -74,7 +74,88 @@ def return_num_columns(img):
     label_p_pred = model_classifier.predict(img_in, verbose=0)
     num_col = np.argmax(label_p_pred[0]) + 1
     return num_col
-    
+
+def return_scaled_image(img, num_col, width_early, model_name):
+        if model_name== ("SBB/eynollah-main-regions-aug-rotation" | "SBB/eynollah-main-regions-aug-scaling" | "SBB/eynollah-main-regions-ensembled"):
+        if num_col == 1 and width_early < 1100:
+            img_w_new = 2000
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 2000)
+        elif num_col == 1 and width_early >= 2500:
+            img_w_new = 2000
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 2000)
+        elif num_col == 1 and width_early >= 1100 and width_early < 2500:
+            img_w_new = width_early
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * width_early)
+        elif num_col == 2 and width_early < 2000:
+            img_w_new = 2400
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 2400)
+        elif num_col == 2 and width_early >= 3500:
+            img_w_new = 2400
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 2400)
+        elif num_col == 2 and width_early >= 2000 and width_early < 3500:
+            img_w_new = width_early
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * width_early)
+        elif num_col == 3 and width_early < 2000:
+            img_w_new = 3000
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 3000)
+        elif num_col == 3 and width_early >= 4000:
+            img_w_new = 3000
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 3000)
+        elif num_col == 3 and width_early >= 2000 and width_early < 4000:
+            img_w_new = width_early
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * width_early)
+        elif num_col == 4 and width_early < 2500:
+            img_w_new = 4000
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 4000)
+        elif num_col == 4 and width_early >= 5000:
+            img_w_new = 4000
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 4000)
+        elif num_col == 4 and width_early >= 2500 and width_early < 5000:
+            img_w_new = width_early
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * width_early)
+        elif num_col == 5 and width_early < 3700:
+            img_w_new = 5000
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 5000)
+        elif num_col == 5 and width_early >= 7000:
+            img_w_new = 5000
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 5000)
+        elif num_col == 5 and width_early >= 3700 and width_early < 7000:
+            img_w_new = width_early
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * width_early)
+        elif num_col == 6 and width_early < 4500:
+            img_w_new = 6500  # 5400
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * 6500)
+        else:
+            img_w_new = width_early
+            img_h_new = int(img.shape[0] / float(img.shape[1]) * width_early)
+        img_new = resize_image(img, img_h_new, img_w_new)
+    if model_name=="SBB/eynollah-main-regions":
+        if num_col == 1:
+            img_w_new = 1000
+            img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
+            
+        elif num_col == 2:
+            img_w_new = 1500
+            img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
+            
+        elif num_col == 3:
+            img_w_new = 2000
+            img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
+            
+        elif num_col == 4:
+            img_w_new = 2500
+            img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
+        elif num_col == 5:
+            img_w_new = 3000
+            img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
+        else:
+            img_w_new = 4000
+            img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
+        img_resized = resize_image(img,img_h_new, img_w_new )
+
+        img = otsu_copy_binary(img_resized)
+    return img
+        
 def do_prediction(model_name, img):
     img_org = np.copy(img)
     model = from_pretrained_keras(model_name)
@@ -110,33 +191,11 @@ def do_prediction(model_name, img):
             img_width_h = img_org.shape[1]
     
             num_col_classifier = return_num_columns(img)
+            width_early = img.shape[1]
     
             
-            
-            if num_col_classifier == 1:
-                img_w_new = 1000
-                img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
-                
-            elif num_col_classifier == 2:
-                img_w_new = 1500
-                img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
-                
-            elif num_col_classifier == 3:
-                img_w_new = 2000
-                img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
-                
-            elif num_col_classifier == 4:
-                img_w_new = 2500
-                img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
-            elif num_col_classifier == 5:
-                img_w_new = 3000
-                img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
-            else:
-                img_w_new = 4000
-                img_h_new = int(img_org.shape[0] / float(img_org.shape[1]) * img_w_new)
-            img_resized = resize_image(img,img_h_new, img_w_new )
-
-            img = otsu_copy_binary(img_resized)
+            img = return_scaled_image(img, num_col_classifier, width_early, model_name)
+