feat: add test for deconvolution

deblur/deblur.py

@@ -0,0 +1,57 @@
+import numpy as np
+import cv2
+
+
+image = np.array([1, 3, 1, 2, 1, 6, 1], dtype=np.float32)
+kernel = np.array([1, 2, 1], dtype=np.float32) / 4
+
+blurred = np.convolve(image, kernel, mode="same")
+
+print(image)
+print(blurred)
+
+print()
+print("building linalg")
+# https://numpy.org/doc/stable/reference/generated/numpy.linalg.solve.html
+a = []
+b = []
+
+for i in range(len(blurred)):
+    y = blurred[i]
+
+    shift = i - 1
+    equation = np.zeros(len(image))
+    # Calculate valid range in the output array
+    start_eq = max(0, shift)
+    end_eq = min(len(image), shift + len(kernel))
+
+    # Corresponding range in the kernel
+    start_k = start_eq - shift  # how much to cut from the beginning of the kernel
+    end_k = start_k + (end_eq - start_eq)
+    
+
+    # Assign the clipped kernel segment
+    equation[start_eq:end_eq] = kernel[start_k:end_k]
+
+    a.append(equation)
+    b.append(y)
+    goal = image[i]
+    print(f"{i} ({goal}): {y} = {equation}")
+
+print()
+print("deblurring")
+deblurred = np.linalg.solve(a, b)
+print(deblurred)
+
+
+
+def show_matrix(m):
+    # Resize the image to make it visible (e.g., scale up to 200x200 pixels)
+    scaled_image = cv2.resize(m, (200, 200), interpolation=cv2.INTER_NEAREST)
+
+    # Display the image
+    cv2.imshow('Test Matrix', scaled_image)
+    cv2.waitKey(0)
+    cv2.destroyAllWindows()
+
+

secure_pixelation/simple_lama_bindings.py

@@ -65,6 +65,7 @@ class SimpleLama:
         print(f"Using device: {self.device}")
         
         model_path = hf_hub_download("okaris/simple-lama", "big-lama.pt")
+        print(f"using model at {model_path}")
         self.model = torch.jit.load(model_path, map_location=self.device).eval()
 
     def __call__(self, image: np.ndarray, mask: np.ndarray) -> np.ndarray: