Allow to flip or mirror images before computing image features

build_feature_index.py

@@ -2,6 +2,7 @@
 
 import numpy as np
 from generic_utils import load_image
+from PIL import Image as pil_image
 
 from app_id_utils import freeze_app_ids, list_app_ids, app_id_to_image_filename
 from data_utils import get_label_database_filename
@@ -14,7 +15,9 @@
 )
 
 
-def build_feature_index(is_horizontal_banner=False, resolution=None):
+def build_feature_index(
+    is_horizontal_banner=False, resolution=None, apply_flip=False, apply_mirror=False
+):
     pooling = "avg"
     feature_filename = get_label_database_filename(pooling)
 
@@ -44,6 +47,10 @@ def build_feature_index(is_horizontal_banner=False, resolution=None):
 
         image_filename = app_id_to_image_filename(app_id, is_horizontal_banner)
         image = load_image(image_filename, target_size=target_model_size)
+        if apply_flip:
+            image = pil_image.flip(image)
+        if apply_mirror:
+            image = pil_image.mirror(image)
         features = convert_image_to_features(image, model, preprocess=preprocess)
 
         Y_hat[counter, :] = features
@@ -58,4 +65,9 @@ def build_feature_index(is_horizontal_banner=False, resolution=None):
 
 
 if __name__ == "__main__":
-    build_feature_index(is_horizontal_banner=False, resolution=None)
+    build_feature_index(
+        is_horizontal_banner=False,
+        resolution=None,
+        apply_flip=False,
+        apply_mirror=False,
+    )