update pre-check function (noise)

Author: ShengkaiXu

evaluation_function/evaluation.py

@@ -41,7 +41,7 @@
             "- brightness: {brightness_mean}\n"
             "- contrast: {contrast_std}\n"
             "- sharpness: {sharpness_score}\n"
-            "- lighting uniformity: {illumination_nonuniformity}\n"
+            "- noise: {noise_score}\n"
             "These values are provided as reference only and do not determine whether the assembly is correct."
         ),
     },
@@ -532,7 +532,7 @@ def qfmt(key: str) -> str:
             brightness_mean=qfmt("brightness_mean"),
             contrast_std=qfmt("contrast_std"),
             sharpness_score=qfmt("sharpness_score"),
-            illumination_nonuniformity=qfmt("illumination_nonuniformity"),
+            noise_score=qfmt("noise_score"),
         )
 
     if task == "single_stage":

evaluation_function/yolo_pipeline.py

@@ -717,26 +717,15 @@ def compute_image_quality_metrics(img_bgr: np.ndarray) -> Dict[str, float]:
     brightness_mean = float(np.mean(gray))
     contrast_std = float(np.std(gray))
     sharpness_score = float(cv2.Laplacian(gray, cv2.CV_64F).var())
-
-    h, w = gray.shape
-    block_means: List[float] = []
-    for r in range(4):
-        for c in range(4):
-            y0 = int(r * h / 4)
-            y1 = int((r + 1) * h / 4)
-            x0 = int(c * w / 4)
-            x1 = int((c + 1) * w / 4)
-            patch = gray[y0:y1, x0:x1]
-            if patch.size > 0:
-                block_means.append(float(np.mean(patch)))
-
-    illumination_nonuniformity = float(np.std(block_means)) if block_means else 0.0
+    blur = cv2.GaussianBlur(gray, (5, 5), 0)
+    noise_residual = gray.astype(np.float32) - blur.astype(np.float32)
+    noise_score = float(np.std(noise_residual))
 
     return {
         "brightness_mean": brightness_mean,
         "contrast_std": contrast_std,
         "sharpness_score": sharpness_score,
-        "illumination_nonuniformity": illumination_nonuniformity,
+        "noise_score": noise_score,
     }