feat: apply mask on top of the image

app.py

@@ -1,6 +1,7 @@
 import gradio
 import numpy
 
+from matplotlib import _cm
 from pathlib import Path
 from PIL import Image
 
@@ -14,29 +15,24 @@ TEST_IMAGES_PATH = Path('.') / 'test'
 def preprocess_mask(file_name):
     """Ensures masks are in grayscale format and removes transparency."""
     mask_path = Path(
-        '/kaggle/input/car-segmentation/car-segmentation/masks') / file_name.name
+        '/kaggle/inumpyut/car-segmentation/car-segmentation/masks') / file_name.name
     mask = Image.open(mask_path)
 
-    # Convert palette-based images to RGBA first to ensure proper color interpretation
     if mask.mode == 'P':
         mask = mask.convert('RGBA')
 
-    # Convert any non-RGBA images to RGBA
     if mask.mode != 'RGBA':
         mask = mask.convert('RGBA')
 
     mask_data = mask.getdata()
 
-    # Replace fully transparent pixels with black (or another valid label)
     new_mask_data = [
-        # Ensure full opacity in new mask
         (r, g, b, 255) if a > 0 else (0, 0, 0, 255)
         for r, g, b, a in mask_data
     ]
 
     mask.putdata(new_mask_data)
 
-    # Convert to grayscale after handling transparency
     return PILMask.create(mask.convert('L'))
 
 
@@ -47,17 +43,27 @@ def segment_image(image):
     image = PILImage.create(image)
     prediction, _, _ = LEARNER.predict(image)
 
+    print("Prediction shape:", prediction.shape)
+    print("Unique values:", numpy.unique(prediction))
+
     prediction_array = numpy.array(prediction, dtype=numpy.uint8)
 
-    # Normalize class indices to 0-255 for proper visualization
-    prediction_array = (prediction_array / prediction_array.max()) * 255
+    colormap = _cm.get_cmap('jet', numpy.max(
+        prediction_array) + 1)
+    colored_mask = colormap(prediction_array)[:, :, :3]
+
+    image_array = numpy.array(image).astype(numpy.float32) / 255.0
+
+    overlay = (image_array * 0.7) + (colored_mask * 0.3)  # Adjust transparency
+
+    overlay = (overlay * 255).astype(numpy.uint8)
 
-    return prediction_array.astype(numpy.uint8)
+    return overlay
 
 
 demo = gradio.Interface(
     segment_image,
-    inputs=gradio.Image(type='pil'),
+    inumpyuts=gradio.Image(type='pil'),
     outputs=gradio.Image(type='numpy'),
     examples=[str(image) for image in TEST_IMAGES_PATH.iterdir()]
 )