Upload app.py

app.py

@@ -0,0 +1,80 @@
+import cv2
+import numpy as np
+import gradio as gr
+
+# Helper function to detect sky condition and get the HSV range
+def detect_sky_color(hsv_image):
+    # Crop the image to the upper half, because we assume the sky is always on the upper half of the image
+    height = hsv_image.shape[0]
+    upper_half_image = hsv_image[:height//2, :]
+
+    # Define color ranges in HSV
+    blue_lower = np.array([46, 17, 148], np.uint8)
+    blue_upper = np.array([154, 185, 249], np.uint8)
+    orange_lower = np.array([10, 100, 100], np.uint8)
+    orange_upper = np.array([25, 183, 254], np.uint8)
+    pale_lower = np.array([0, 0, 129], np.uint8)
+    pale_upper = np.array([171, 64, 225], np.uint8)
+
+    # Create masks for colors
+    blue_mask = cv2.inRange(upper_half_image, blue_lower, blue_upper)
+    orange_mask = cv2.inRange(upper_half_image, orange_lower, orange_upper)
+    pale_mask = cv2.inRange(upper_half_image, pale_lower, pale_upper)
+
+    # Calculate the percentage of cropped image covered by each color
+    blue_percentage = np.sum(blue_mask > 0) / (upper_half_image.shape[0] * upper_half_image.shape[1]) * 100
+    orange_percentage = np.sum(orange_mask > 0) / (upper_half_image.shape[0] * upper_half_image.shape[1]) * 100
+    pale_percentage = np.sum(pale_mask > 0) / (upper_half_image.shape[0] * upper_half_image.shape[1]) * 100
+
+    # Determine the predominant color in the upper half
+    max_color = max(blue_percentage, orange_percentage, pale_percentage)
+    if max_color == blue_percentage:
+        return blue_lower, blue_upper
+    elif max_color == orange_percentage:
+        return orange_lower, orange_upper
+    else:
+        return pale_lower, pale_upper
+
+
+# Main function to process image and display sky masks
+def sky_segmentation(uploaded_image):
+    # Read the image
+    image = cv2.imread(uploaded_image)
+
+    # Convert to HSV image
+    hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
+
+    # Determine HSV range based on helper function
+    (hsv_lower, hsv_upper) = detect_sky_color(hsv)
+
+    # Use hsv_lower and hsv_upper to create a mask, which isolates the sky region
+    mask_initial = cv2.inRange(hsv, hsv_lower, hsv_upper)
+
+    # Apply morphological operations to fine-tune the mask
+    kernel = np.ones((3,3), np.uint8)
+    mask_fine_tuned = cv2.erode(mask_initial, kernel, iterations=1)
+    mask_fine_tuned = cv2.dilate(mask_fine_tuned, kernel, iterations=1)
+
+    # Perform connected component analysis
+    num_labels, labels_im = cv2.connectedComponents(mask_fine_tuned)
+
+    # Create an array to hold the size of each component
+    sizes = np.bincount(labels_im.flatten())
+
+    # Set the size of the background (label 0) to zero
+    sizes[0] = 0
+
+    # Find the largest component
+    max_label = np.argmax(sizes)
+
+    # Create a mask with only the largest component
+    sky_mask = np.zeros_like(mask_fine_tuned)
+    sky_mask[labels_im == max_label] = 255    
+    
+    return sky_mask
+
+
+# Create a Gradio demo
+demo = gr.Interface(sky_segmentation, gr.Image(type='filepath'), "image")
+if __name__ == "__main__":
+    demo.launch(share=True)