Update app.py

app.py

@@ -136,9 +136,10 @@ def expand_mask(mask, kernel_size):
     return Image.fromarray(expanded_mask_array)
 
 
-def crop_face_to_square(image_rgb, padding_ratio=0.2):
+def crop_face_to_rectangle_with_body(image_rgb, padding_ratio=0.2, height_multiplier=1.5):
     """
-    Detects the face in the input image and crops an enlarged square region around it.
+    Detect the face and crop a rectangular region that includes more of the body below the face.
+    Instead of centering around the face, we start near the face region and extend downward.
     """
     face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')
     gray_image = cv2.cvtColor(image_rgb, cv2.COLOR_RGB2GRAY)
@@ -149,22 +150,29 @@ def crop_face_to_square(image_rgb, padding_ratio=0.2):
         return None
 
     x, y, w, h = faces[0]
-    center_x, center_y = x + w // 2, y + h // 2
-    side_length = max(w, h)
-    padded_side_length = int(side_length * (1 + padding_ratio))
-    half_side = padded_side_length // 2
+    face_x_center = x + w // 2
+    face_y_top = y 
+    face_side_length = max(w, h)
+    padded_side_length = int(face_side_length * (1 + padding_ratio))
 
-    top_left_x = max(center_x - half_side, 0)
-    top_left_y = max(center_y - half_side, 0)
-    bottom_right_x = min(center_x + half_side, image_rgb.shape[1])
-    bottom_right_y = min(center_y + half_side, image_rgb.shape[0])
+    cropped_width = padded_side_length
+    cropped_height = int(padded_side_length * height_multiplier)
+
+    top_left_x = max(face_x_center - cropped_width // 2, 0)
+
+    top_margin = int(padded_side_length * 0.1)
+    top_left_y = max(face_y_top - top_margin, 0)
+    bottom_right_x = min(top_left_x + cropped_width, image_rgb.shape[1])
+    bottom_right_y = min(top_left_y + cropped_height, image_rgb.shape[0])
 
     cropped_image = image_rgb[top_left_y:bottom_right_y, top_left_x:bottom_right_x]
-    resized_image = cv2.resize(cropped_image, (768, 768), interpolation=cv2.INTER_AREA)
+
+    resized_image = cv2.resize(cropped_image, (768, 1024), interpolation=cv2.INTER_AREA)
 
     return resized_image
 
 
+
 def spirit_animal_baseline(image_path, num_images = 4):
 
     image = cv2.imread(image_path)
@@ -689,7 +697,7 @@ with gr.Blocks() as demo:
                     background_option = gr.Radio(choices=["Preserve Background", "Don't Preserve Background"], label="Background Option", value="Preserve Background")
                     generate_image_button = gr.Button("Generate Image")
                     gr.Examples(
-                        examples=["example1.jpg", "example2.jpg", "example3.jpg"],
+                        examples=["example1.jpg", "example2.jpg", "example3.jpg", "example4.jpg"],
                         inputs=image_input,
                         label="Example Images"
                     )