feat: Enable comprehensive mypy type checking across landmarkdiff codebase

.github/workflows/ci.yml

@@ -29,6 +29,13 @@ jobs:
   type-check:
     runs-on: ubuntu-latest
     needs: lint
+    strategy:
+      fail-fast: false
+      matrix:
+        target:
+          - landmarkdiff/
+          - scripts/
+          - benchmarks/
     steps:
       - uses: actions/checkout@v5
       - uses: actions/setup-python@v6
@@ -37,8 +44,8 @@ jobs:
           cache: pip
       - name: Install dependencies
         run: pip install -e ".[dev]"
-      - name: Type check
-        run: mypy landmarkdiff/ --ignore-missing-imports
+      - name: Type check ${{ matrix.target }}
+        run: mypy ${{ matrix.target }}
   pre-commit:
     runs-on: ubuntu-latest
     needs: lint

.pre-commit-config.yaml

@@ -28,6 +28,7 @@ repos:
         additional_dependencies: [types-Pillow, types-PyYAML, types-requests]
         exclude: '\.ipynb$'
         args: [
+          --explicit-package-bases,
           --ignore-missing-imports,
           --config-file=pyproject.toml,
           --exclude, 'scripts/benchmark_inference\.py',

benchmarks/__init__.py

@@ -0,0 +1 @@
+"""Benchmarks package for LandmarkDiff performance testing."""

landmarkdiff/api_client_async.py

@@ -22,7 +22,7 @@ async def main():
 import base64
 import logging
 from pathlib import Path
-from typing import Any
+from typing import Any, cast
 
 import cv2
 import numpy as np
@@ -91,7 +91,7 @@ async def _handle_response(self, resp: Any) -> dict[str, Any]:
         if resp.status >= 400:
             text = await resp.text()
             raise LandmarkDiffAPIError(f"Server returned {resp.status}: {text[:200]}")
-        return await resp.json()
+        return cast(dict[str, Any], await resp.json())
 
     # ------------------------------------------------------------------
     # API methods
@@ -124,7 +124,7 @@ async def procedures(self) -> list[str]:
         try:
             async with session.get(f"{self.base_url}/procedures") as resp:
                 data = await self._handle_response(resp)
-                return data.get("procedures", [])
+                return cast(list[str], data.get("procedures", []))
         except LandmarkDiffAPIError:
             raise
         except Exception as e:

landmarkdiff/comparison.py

@@ -7,6 +7,7 @@
 from __future__ import annotations
 
 import logging
+from typing import cast
 
 import cv2
 import numpy as np
@@ -47,7 +48,7 @@ def create_slider_composite(
     if line_width > 0 and 0 < split_x < w:
         cv2.line(result, (split_x, 0), (split_x, h - 1), line_color, line_width)
 
-    return result
+    return cast(np.ndarray, result)
 
 
 def create_side_by_side(
@@ -101,7 +102,7 @@ def create_side_by_side(
                 cv2.LINE_AA,
             )
 
-    return canvas
+    return cast(np.ndarray, canvas)
 
 
 def create_difference_heatmap(
@@ -127,7 +128,7 @@ def create_difference_heatmap(
     diff = cv2.absdiff(original, pred)
     gray_diff = cv2.cvtColor(diff, cv2.COLOR_BGR2GRAY)
     amplified = np.clip(gray_diff.astype(np.float32) * amplify, 0, 255).astype(np.uint8)
-    return cv2.applyColorMap(amplified, colormap)
+    return cast(np.ndarray, cv2.applyColorMap(amplified, colormap))
 
 
 def create_checkerboard_blend(
@@ -166,4 +167,4 @@ def create_checkerboard_blend(
 
     mask_3ch = np.stack([mask] * 3, axis=-1)
     result = original.astype(np.float32) * mask_3ch + pred.astype(np.float32) * (1.0 - mask_3ch)
-    return np.clip(result, 0, 255).astype(np.uint8)
+    return cast(np.ndarray, np.clip(result, 0, 255).astype(np.uint8))

landmarkdiff/planning.py

@@ -8,6 +8,7 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
+from typing import cast
 
 import cv2
 import numpy as np
@@ -214,4 +215,4 @@ def visualize_planning(
             cv2.LINE_AA,
         )
 
-    return canvas
+    return cast(np.ndarray, canvas)

pyproject.toml

@@ -192,16 +192,19 @@ ignore = [
 
 [tool.mypy]
 python_version = "3.10"
-warn_return_any = false
+warn_return_any = true
 warn_unused_configs = true
-disallow_untyped_defs = false
-check_untyped_defs = false
+disallow_untyped_defs = true
+check_untyped_defs = true
 explicit_package_bases = true
 ignore_missing_imports = true
+strict_optional = true
+warn_redundant_casts = true
+warn_unused_ignores = true
 exclude = [
-    "scripts/batch_inference.py",
-    "scripts/benchmark_inference.py",
     "examples/*.ipynb",
+    "tests/*.py",
+    "examples/",
 ]
 
 [[tool.mypy.overrides]]
@@ -246,3 +249,11 @@ ignore_errors = true
 [[tool.mypy.overrides]]
 module = ["scripts.*"]
 ignore_errors = true
+
+[[tool.mypy.overrides]]
+module = ["tests.*"]
+ignore_errors = true
+
+[[tool.mypy.overrides]]
+module = ["examples.*"]
+ignore_errors = true

scripts/__init__.py

@@ -0,0 +1 @@
+"""Scripts package for LandmarkDiff utilities and experiments."""