removing naize harmonization, adding functions for creating masks

app.py

@@ -185,9 +185,10 @@ def start_tryon(human_img_dict,garm_img,garment_des, background_img, is_checked,
     else:
         human_img = human_img_orig.resize((WIDTH, HEIGHT))
 
+    # Commenting out naize harmonization for now. We will have to integrate with Deep Learning based Harmonization methods
     # Do color transfer from background image for better image harmonization
-    if background_img:
-        human_img = BackgroundProcessor.intensity_transfer(human_img, background_img)
+    #if background_img:
+    #    human_img = BackgroundProcessor.intensity_transfer(human_img, background_img)
 
 
     if is_checked:

src/background_processor.py

@@ -232,6 +232,63 @@ class BackgroundProcessor:
         logging.error(f"failed to use remove bg. Status: {remove_bg_request.status_code}. Resp: {remove_bg_request.content}")
         return None
     
+    @classmethod
+    def create_mask(cls, foreground_path: str, mask_path: str):
+        """
+        Given foreground image path with background removed, create a maska and save it in mask_path
+        """
+        # Load the foreground image with alpha channel
+        foreground = Image.open(foreground_path)
+
+        # Convert to RGBA if not already
+        foreground = foreground.convert("RGBA")
+
+        # Create the mask from the alpha channel
+        alpha_channel = np.array(foreground.split()[-1])
+
+        # Create a binary mask where alpha > 0 is white (255) and alpha == 0 is black (0)
+        mask = np.where(alpha_channel > 0, 255, 0).astype(np.uint8)
+
+        # Save the mask to a file
+        Image.fromarray(mask).save(mask_path) 
+    
+    @classmethod
+    def get_minimal_bounding_box(cls, foreground_pil: Image):
+        """
+        Result x1,y1,x2,y2 ie cordinate of bottom left and top right
+        """
+        # convert to cv2
+        foreground = ImageFormatConvertor.pil_to_cv2(foreground_pil)
+        # Ensure the image has an alpha channel (transparency)
+        if foreground.shape[2] == 4:
+            # Extract the alpha channel
+            alpha_channel = foreground[:, :, 3]            
+            # Create a binary image from the alpha channel
+            _, binary_image = cv2.threshold(alpha_channel, 1, 255, cv2.THRESH_BINARY)
+        else:
+            # If there is no alpha channel, convert the image to grayscale
+            gray_image = cv2.cvtColor(foreground, cv2.COLOR_BGR2GRAY)            
+            # Apply binary thresholding
+            _, binary_image = cv2.threshold(gray_image, 1, 255, cv2.THRESH_BINARY)
+        
+        # Find all non-zero points (non-background)
+        non_zero_points = cv2.findNonZero(binary_image)
+        
+        # Get the minimal bounding rectangle
+        if non_zero_points is not None:
+            x, y, w, h = cv2.boundingRect(non_zero_points)
+            """
+            # Optionally, draw the bounding box on the image for visualization
+            output_image = foreground.copy()
+            cv2.rectangle(output_image, (x, y), (x+w, y+h), (0, 255, 0, 255), 2)
+            # Save or display the output image
+            output_image_pil = ImageFormatConvertor.cv2_to_pil(output_image)
+            output_image_pil.save('output_with_bounding_box.png')
+            """
+            
+            return (x, y, x + w, y + h)
+        else:
+            return 0,0,w,h
     
     @classmethod
     def color_transfer(cls, source_pil: Image, target_pil: Image) -> Image: