Resolve paths to images

src/generate_emr_csv.py

@@ -145,7 +145,7 @@ def generate_dataset(image_dir_override=None, output_path_override=None):
 
         for i in range(SAMPLES_PER_CLASS):
             image_path = str(
-                random.choice(image_files).relative_to(root_image_dir.parent)  # path of image respective to the project root
+                random.choice(image_files).relative_to(CURRENT_DIR.parent)  # path of image respective to the project root
             )
             text = build_emr(label, i)
             triage = triage_map[label]

src/triage_dataset.py

@@ -89,14 +89,11 @@ class TriageDataset(Dataset):
             # Process image
             image_path = Path(row["image_path"])
 
-            # Normalize if CSV path includes redundant 'data/images/' prefix
-            redundant_prefix = Path("data/images")
-            if image_path.parts[:2] == redundant_prefix.parts:
-                image_path = Path(*image_path.parts[2:])  # drop 'data/images'
-
-            # Combine with base directory if not absolute
             if not image_path.is_absolute():
-                image_path = self.image_base_dir / image_path
+                if image_path.parts[:2] == ("data", self.image_base_dir.name):
+                    image_path = self.image_base_dir.parent / Path(*image_path.parts[1:])
+                else:
+                    image_path = self.image_base_dir / image_path
 
             if not image_path.exists():
                 if IS_CI:

tests/test_generate_emr_csv.py

@@ -115,7 +115,7 @@ def test_emr_text_quality():
 
 
 def test_image_path_format():
-    expected_path = DUMMY_IMAGES_DIR.relative_to(DUMMY_IMAGES_DIR.parent) if IS_CI else REAL_IMAGES_DIR.relative_to(REAL_IMAGES_DIR.parent)
+    expected_path = (DATA_DIR / ("dummy_images" if IS_CI else "images")).relative_to(BASE_DIR)
     with open(CSV_PATH, newline="") as f:
         reader = csv.DictReader(f)
         for row in reader: