adding tiling processor

src/controlnet_aux/__init__.py

@@ -14,3 +14,4 @@
 from .shuffle import ContentShuffleDetector
 from .mediapipe_face import MediapipeFaceDetector
 from .segment_anything import SamDetector
+from .tiling import TilingDetector

src/controlnet_aux/processor.py

@@ -18,7 +18,9 @@
                             CannyDetector,
                             ContentShuffleDetector,
                             ZoeDetector,
-                            MediapipeFaceDetector)
+                            MediapipeFaceDetector,
+                            TilingDetector,
+                            )
 
 LOGGER = logging.getLogger(__name__)
 
@@ -44,11 +46,12 @@
     'lineart_coarse': {'class': LineartDetector, 'checkpoint': True},
     'lineart_realistic': {'class': LineartDetector, 'checkpoint': True},
     'lineart_anime': {'class': LineartAnimeDetector, 'checkpoint': True},
-    'depth_zoe': {'class': ZoeDetector, 'checkpoint': True}, 
+    'depth_zoe': {'class': ZoeDetector, 'checkpoint': True},
     # instantiate
     'shuffle': {'class': ContentShuffleDetector, 'checkpoint': False},
     'mediapipe_face': {'class': MediapipeFaceDetector, 'checkpoint': False},
     'canny': {'class': CannyDetector, 'checkpoint': False},
+    'tiling': {'class': TilingDetector, 'checkpoint': False},
 }
 
 
@@ -76,6 +79,7 @@
     'shuffle': {'resize': False},
     'depth_zoe': {'resize': False},
     'mediapipe_face': {'resize': False},
+    'tiling': {'resize': False},
 }
 
 CHOICES = f"Choices for the processor are {list(MODELS.keys())}"

src/controlnet_aux/tiling/__init__.py

@@ -0,0 +1,25 @@
+from PIL import Image
+
+
+class TilingDetector:
+    def __call__(self, input_image: Image.Image, resolution: int = 1024):
+        input_image = input_image.convert("RGB")
+        original_width, original_height = input_image.size
+
+        # Calculate the scale factor based on the desired resolution
+        scale_factor = float(resolution) / min(original_height, original_width)
+
+        # Scale the width and height of the image
+        scaled_height = original_height * scale_factor
+        scaled_width = original_width * scale_factor
+
+        # Round the scaled dimensions to the nearest multiple of 64
+        scaled_height = int(round(scaled_height / 64.0)) * 64
+        scaled_width = int(round(scaled_width / 64.0)) * 64
+
+        # Resize the image using Lanczos interpolation
+        resized_image = input_image.resize(
+            (scaled_width, scaled_height), resample=Image.LANCZOS
+        )
+
+        return resized_image