dragon-ml-toolbox 14.7.0__py3-none-any.whl → 16.2.1__py3-none-any.whl

ml_tools/ML_utilities.py

@@ -7,7 +7,7 @@ from torch import nn
 from .path_manager import make_fullpath, list_subdirectories, list_files_by_extension
 from ._script_info import _script_info
 from ._logger import _LOGGER
-from .keys import DatasetKeys, PytorchModelArchitectureKeys, PytorchArtifactPathKeys, SHAPKeys, UtilityKeys, PyTorchCheckpointKeys
+from ._keys import DatasetKeys, PytorchModelArchitectureKeys, PytorchArtifactPathKeys, SHAPKeys, UtilityKeys, PyTorchCheckpointKeys
 from .utilities import load_dataframe
 from .custom_logger import save_list_strings, custom_logger
 from .serde import serialize_object_filename

ml_tools/ML_vision_datasetmaster.py

@@ -12,25 +12,23 @@ import random
 import json
 import inspect
 
-from .ML_datasetmaster import _BaseMaker
 from .path_manager import make_fullpath
 from ._logger import _LOGGER
 from ._script_info import _script_info
-from .keys import VisionTransformRecipeKeys, ObjectDetectionKeys
-from ._ML_vision_recipe import save_recipe
-from .ML_vision_transformers import TRANSFORM_REGISTRY
+from ._keys import VisionTransformRecipeKeys, ObjectDetectionKeys
+from .ML_vision_transformers import TRANSFORM_REGISTRY, _save_recipe
 from .custom_logger import custom_logger
 
 
 __all__ = [
-    "VisionDatasetMaker",
-    "SegmentationDatasetMaker",
-    "ObjectDetectionDatasetMaker"
+    "DragonDatasetVision",
+    "DragonDatasetSegmentation",
+    "DragonDatasetObjectDetection"
 ]
 
 
-# --- VisionDatasetMaker ---
-class VisionDatasetMaker(_BaseMaker):
+# --- Vision Maker ---
+class DragonDatasetVision:
     """
     Creates processed PyTorch datasets for computer vision tasks from an
     image folder directory.
@@ -45,10 +43,12 @@ class VisionDatasetMaker(_BaseMaker):
         """
         Typically not called directly. Use the class methods `from_folder()` or `from_folders()` to create an instance.
         """
-        super().__init__()
+        self._train_dataset = None
+        self._test_dataset = None
+        self._val_dataset = None
         self._full_dataset: Optional[ImageFolder] = None
         self.labels: Optional[List[int]] = None
-        self.class_map: Optional[dict[str,int]] = None
+        self.class_map: dict[str,int] = dict()
         
         self._is_split = False
         self._are_transforms_configured = False
@@ -56,7 +56,7 @@ class VisionDatasetMaker(_BaseMaker):
         self._has_mean_std: bool = False
 
     @classmethod
-    def from_folder(cls, root_dir: Union[str,Path]) -> 'VisionDatasetMaker':
+    def from_folder(cls, root_dir: Union[str,Path]) -> 'DragonDatasetVision':
         """
         Creates a maker instance from a single root directory of images.
         
@@ -70,7 +70,7 @@ class VisionDatasetMaker(_BaseMaker):
             root_dir (str | Path): The path to the root directory containing class subfolders.
 
         Returns:
-            VisionDatasetMaker: A new instance with the full dataset loaded.
+            Instance: A new instance with the full dataset loaded.
         """
         root_path = make_fullpath(root_dir, enforce="directory")
         # Load with NO transform. We get PIL Images.
@@ -87,7 +87,7 @@ class VisionDatasetMaker(_BaseMaker):
     def from_folders(cls, 
                      train_dir: Union[str,Path], 
                      val_dir: Union[str,Path], 
-                     test_dir: Optional[Union[str,Path]] = None) -> 'VisionDatasetMaker':
+                     test_dir: Optional[Union[str,Path]] = None) -> 'DragonDatasetVision':
         """
         Creates a maker instance from separate, pre-split directories.
         
@@ -101,7 +101,7 @@ class VisionDatasetMaker(_BaseMaker):
             test_dir (str | Path | None): Path to the test data directory.
 
         Returns:
-            VisionDatasetMaker: A new, pre-split instance.
+            Instance: A new, pre-split instance.
 
         Raises:
             ValueError: If the classes found in train, val, or test directories are inconsistent.
@@ -186,7 +186,7 @@ class VisionDatasetMaker(_BaseMaker):
         print(report)
 
     def split_data(self, val_size: float = 0.2, test_size: float = 0.0, 
-                   stratify: bool = True, random_state: Optional[int] = None) -> 'VisionDatasetMaker':
+                   stratify: bool = True, random_state: Optional[int] = None) -> 'DragonDatasetVision':
         """
         Splits the dataset into train, validation, and optional test sets.
         
@@ -202,7 +202,7 @@ class VisionDatasetMaker(_BaseMaker):
             random_state (int | None): Seed for the random number generator for reproducible splits.
 
         Returns:
-            VisionDatasetMaker: The same instance, now with datasets split.
+            Self: The same instance, now with datasets split.
             
         Raises:
             ValueError: If `val_size` and `test_size` sum to 1.0 or more.
@@ -249,22 +249,23 @@ class VisionDatasetMaker(_BaseMaker):
         _LOGGER.info(f"Data split into: \n- Training: {len(self._train_dataset)} images \n- Validation: {len(self._val_dataset)} images")
         return self
 
-    def configure_transforms(self, resize_size: int = 256, crop_size: int = 224, 
+    def configure_transforms(self, 
+                             resize_size: int = 256, 
+                             crop_size: int = 224, 
                              mean: Optional[List[float]] = [0.485, 0.456, 0.406], 
                              std: Optional[List[float]] = [0.229, 0.224, 0.225],
                              pre_transforms: Optional[List[Callable]] = None,
-                             extra_train_transforms: Optional[List[Callable]] = None) -> 'VisionDatasetMaker':
+                             extra_train_transforms: Optional[List[Callable]] = None) -> 'DragonDatasetVision':
         """
         Configures and applies the image transformations and augmentations.
         
         This method must be called AFTER data is loaded and split.
         
         It sets up two pipelines:
-        1.  **Training Pipeline:** Includes random augmentations like
-            `RandomResizedCrop` and `RandomHorizontalFlip` (plus any
+        1.  **Training Pipeline:** Includes random augmentations:
+            `RandomResizedCrop(crop_size)`, `RandomHorizontalFlip(0.5)`, and `RandomRotation(90)` (plus any
             `extra_train_transforms`) for online augmentation.
-        2.  **Validation/Test Pipeline:** A deterministic pipeline using
-            `Resize` and `CenterCrop` for consistent evaluation.
+        2.  **Validation/Test Pipeline:** A deterministic pipeline using `Resize` and `CenterCrop` for consistent evaluation.
             
         Both pipelines finish with `ToTensor` and `Normalize`.
 
@@ -279,7 +280,7 @@ class VisionDatasetMaker(_BaseMaker):
             pre_transforms (List[Callable] | None): An list of transforms to be applied at the very beginning of the transformations for all sets.
 
         Returns:
-            VisionDatasetMaker: The same instance, with transforms applied.
+            Self: The same instance, with transforms applied.
             
         Raises:
             RuntimeError: If called before data is split.
@@ -315,8 +316,9 @@ class VisionDatasetMaker(_BaseMaker):
 
         # Base augmentations for training
         base_train_transforms = [
-            transforms.RandomResizedCrop(crop_size), 
-            transforms.RandomHorizontalFlip()
+            transforms.RandomResizedCrop(size=crop_size), 
+            transforms.RandomHorizontalFlip(p=0.5),
+            transforms.RandomRotation(degrees=90)
         ]
         if extra_train_transforms:
             base_train_transforms.extend(extra_train_transforms)
@@ -350,10 +352,10 @@ class VisionDatasetMaker(_BaseMaker):
         # --- Apply Transforms using the Wrapper ---
         # This correctly assigns the transform regardless of whether the dataset is a Subset (from_folder) or an ImageFolder (from_folders).
         
-        self._train_dataset = _DatasetTransformer(self._train_dataset, train_transform) # type: ignore
-        self._val_dataset = _DatasetTransformer(self._val_dataset, val_transform) # type: ignore
+        self._train_dataset = _DatasetTransformer(self._train_dataset, train_transform, self.class_map) # type: ignore
+        self._val_dataset = _DatasetTransformer(self._val_dataset, val_transform, self.class_map) # type: ignore
         if self._test_dataset:
-            self._test_dataset = _DatasetTransformer(self._test_dataset, val_transform) # type: ignore
+            self._test_dataset = _DatasetTransformer(self._test_dataset, val_transform, self.class_map) # type: ignore
         
         self._are_transforms_configured = True
         _LOGGER.info("Image transforms configured and applied.")
@@ -381,8 +383,8 @@ class VisionDatasetMaker(_BaseMaker):
             _LOGGER.warning("Transforms have not been configured.")
 
         if self._test_dataset:
-            return self._train_dataset, self._val_dataset, self._test_dataset
-        return self._train_dataset, self._val_dataset
+            return self._train_dataset, self._val_dataset, self._test_dataset # type: ignore
+        return self._train_dataset, self._val_dataset # type: ignore
 
     def save_transform_recipe(self, filepath: Union[str, Path]) -> None:
         """
@@ -481,7 +483,7 @@ class VisionDatasetMaker(_BaseMaker):
             )
         
         # 3. Save the file
-        save_recipe(recipe, file_path)
+        _save_recipe(recipe, file_path)
         
     def save_class_map(self, save_dir: Union[str,Path]) -> dict[str,int]:
         """
@@ -499,6 +501,21 @@ class VisionDatasetMaker(_BaseMaker):
         
         return self.class_map
     
+    def images_per_dataset(self) -> str:
+        """
+        Get the number of images per dataset as a string.
+        """
+        if self._is_split:
+            train_len = len(self._train_dataset) if self._train_dataset else 0
+            val_len = len(self._val_dataset) if self._val_dataset else 0
+            test_len = len(self._test_dataset) if self._test_dataset else 0
+            return f"Train | Validation | Test: {train_len} | {val_len} | {test_len} images"
+        elif self._full_dataset:
+            return f"Full Dataset: {len(self._full_dataset)} images"
+        else:
+            _LOGGER.warning("No datasets found.")
+            return "No datasets found"   
+    
     def __repr__(self) -> str:
         s = f"<{self.__class__.__name__}>:\n"
         s += f"  Split: {self._is_split}\n"
@@ -511,7 +528,7 @@ class VisionDatasetMaker(_BaseMaker):
             train_len = len(self._train_dataset) if self._train_dataset else 0
             val_len = len(self._val_dataset) if self._val_dataset else 0
             test_len = len(self._test_dataset) if self._test_dataset else 0
-            s += f"  Datasets (Train/Val/Test): {train_len} / {val_len} / {test_len}\n"
+            s += f"  Datasets (Train|Val|Test): {train_len} | {val_len} | {test_len}\n"
         elif self._full_dataset:
             s += f"  Full Dataset Size: {len(self._full_dataset)} images\n"
             
@@ -523,9 +540,10 @@ class _DatasetTransformer(Dataset):
     Internal wrapper class to apply a specific transform pipeline to any
     dataset (e.g., a full ImageFolder or a Subset).
     """
-    def __init__(self, dataset: Dataset, transform: Optional[transforms.Compose] = None):
+    def __init__(self, dataset: Dataset, transform: Optional[transforms.Compose] = None, class_map: dict[str,int]=dict()):
         self.dataset = dataset
         self.transform = transform
+        self.class_map = class_map
         
         # --- Propagate attributes for inspection ---
         # For ImageFolder
@@ -565,7 +583,7 @@ class _SegmentationDataset(Dataset):
         self.mask_paths = mask_paths
         self.transform = transform
         
-        # --- Propagate 'classes' if they exist (for MLTrainer) ---
+        # --- Propagate 'classes' if they exist for trainer ---
         self.classes: List[str] = []
 
     def __len__(self):
@@ -673,8 +691,8 @@ class _PairedRandomResizedCrop:
         
         return cropped_image, cropped_mask # type: ignore
 
-# --- SegmentationDatasetMaker ---
-class SegmentationDatasetMaker(_BaseMaker):
+# --- Segmentation Dataset ---
+class DragonDatasetSegmentation:
     """
     Creates processed PyTorch datasets for segmentation from image and mask folders.
 
@@ -683,7 +701,7 @@ class SegmentationDatasetMaker(_BaseMaker):
     to both the image and its corresponding mask.
     
     Workflow:
-    1. `maker = SegmentationDatasetMaker.from_folders(img_dir, mask_dir)`
+    1. `maker = DragonDatasetSegmentation.from_folders(img_dir, mask_dir)`
     2. `maker.set_class_map({'background': 0, 'road': 1})`
     3. `maker.split_data(val_size=0.2)`
     4. `maker.configure_transforms(crop_size=256)`
@@ -695,7 +713,9 @@ class SegmentationDatasetMaker(_BaseMaker):
         """
         Typically not called directly. Use the class method `from_folders()` to create an instance.
         """
-        super().__init__()
+        self._train_dataset = None
+        self._test_dataset = None
+        self._val_dataset = None
         self.image_paths: List[Path] = []
         self.mask_paths: List[Path] = []
         self.class_map: Dict[str, int] = {}
@@ -707,7 +727,7 @@ class SegmentationDatasetMaker(_BaseMaker):
         self._has_mean_std: bool = False
 
     @classmethod
-    def from_folders(cls, image_dir: Union[str, Path], mask_dir: Union[str, Path]) -> 'SegmentationDatasetMaker':
+    def from_folders(cls, image_dir: Union[str, Path], mask_dir: Union[str, Path]) -> 'DragonDatasetSegmentation':
         """
         Creates a maker instance by loading all matching image-mask pairs
         from two corresponding directories.
@@ -720,7 +740,7 @@ class SegmentationDatasetMaker(_BaseMaker):
             mask_dir (str | Path): Path to the directory containing segmentation masks.
 
         Returns:
-            SegmentationDatasetMaker: A new instance with all pairs loaded.
+            DragonDatasetSegmentation: A new instance with all pairs loaded.
         """
         maker = cls()
         img_path_obj = make_fullpath(image_dir, enforce="directory")
@@ -780,14 +800,14 @@ class SegmentationDatasetMaker(_BaseMaker):
         Logs a report of the types, sizes, and channels of image files
         found in the directory. Useful for checking masks.
         """
-        VisionDatasetMaker.inspect_folder(path)
+        DragonDatasetVision.inspect_folder(path)
 
-    def set_class_map(self, class_map: Dict[str, int]) -> 'SegmentationDatasetMaker':
+    def set_class_map(self, class_map: Dict[str, int]) -> 'DragonDatasetSegmentation':
         """
-        Sets a map of pixel_value -> class_name. This is used by the MLTrainer for clear evaluation reports.
+        Sets a map of class_name -> pixel value. This is used by the Trainer for clear evaluation reports.
 
         Args:
-            class_map (Dict[int, str]): A dictionary mapping the integer pixel
+            class_map (Dict[str, int]): A dictionary mapping the integer pixel
                 value in a mask to its string name.
                 Example: {'background': 0, 'road': 1, 'car': 2}
         """
@@ -803,7 +823,7 @@ class SegmentationDatasetMaker(_BaseMaker):
         return []
 
     def split_data(self, val_size: float = 0.2, test_size: float = 0.0, 
-                   random_state: Optional[int] = 42) -> 'SegmentationDatasetMaker':
+                   random_state: Optional[int] = 42) -> 'DragonDatasetSegmentation':
         """
         Splits the loaded image-mask pairs into train, validation, and test sets.
 
@@ -813,7 +833,7 @@ class SegmentationDatasetMaker(_BaseMaker):
             random_state (int | None): Seed for reproducible splits.
 
         Returns:
-            SegmentationDatasetMaker: The same instance, now with datasets created.
+            DragonDatasetSegmentation: The same instance, now with datasets created.
         """
         if self._is_split:
             _LOGGER.warning("Data has already been split.")
@@ -857,7 +877,7 @@ class SegmentationDatasetMaker(_BaseMaker):
         self._train_dataset = _SegmentationDataset(train_imgs, train_masks, transform=None)
         self._val_dataset = _SegmentationDataset(val_imgs, val_masks, transform=None)
         
-        # Propagate class names to datasets for MLTrainer
+        # Propagate class names to datasets for trainer
         self._train_dataset.classes = self.classes # type: ignore
         self._val_dataset.classes = self.classes # type: ignore
 
@@ -869,7 +889,7 @@ class SegmentationDatasetMaker(_BaseMaker):
                              resize_size: int = 256, 
                              crop_size: int = 224, 
                              mean: Optional[List[float]] = [0.485, 0.456, 0.406], 
-                             std: Optional[List[float]] = [0.229, 0.224, 0.225]) -> 'SegmentationDatasetMaker':
+                             std: Optional[List[float]] = [0.229, 0.224, 0.225]) -> 'DragonDatasetSegmentation':
         """
         Configures and applies the image and mask transformations.
         
@@ -884,7 +904,7 @@ class SegmentationDatasetMaker(_BaseMaker):
             std (List[float] | None): The std dev values for image normalization.
 
         Returns:
-            SegmentationDatasetMaker: The same instance, with transforms applied.
+            DragonDatasetSegmentation: The same instance, with transforms applied.
         """
         if not self._is_split:
             _LOGGER.error("Transforms must be configured AFTER splitting data. Call .split_data() first.")
@@ -961,8 +981,8 @@ class SegmentationDatasetMaker(_BaseMaker):
             raise RuntimeError()
 
         if self._test_dataset:
-            return self._train_dataset, self._val_dataset, self._test_dataset
-        return self._train_dataset, self._val_dataset
+            return self._train_dataset, self._val_dataset, self._test_dataset # type: ignore
+        return self._train_dataset, self._val_dataset # type: ignore
     
     def save_transform_recipe(self, filepath: Union[str, Path]) -> None:
         """
@@ -1006,7 +1026,20 @@ class SegmentationDatasetMaker(_BaseMaker):
             )
         
         # Save the file
-        save_recipe(recipe, file_path)
+        _save_recipe(recipe, file_path)
+        
+    def images_per_dataset(self) -> str:
+        """
+        Get the number of images per dataset as a string.
+        """
+        if self._is_split:
+            train_len = len(self._train_dataset) if self._train_dataset else 0
+            val_len = len(self._val_dataset) if self._val_dataset else 0
+            test_len = len(self._test_dataset) if self._test_dataset else 0
+            return f"Train | Validation | Test: {train_len} | {val_len} | {test_len} images"
+        else:
+            _LOGGER.warning("No datasets found.")
+            return "No datasets found"   
         
     def __repr__(self) -> str:
         s = f"<{self.__class__.__name__}>:\n"
@@ -1021,7 +1054,7 @@ class SegmentationDatasetMaker(_BaseMaker):
             train_len = len(self._train_dataset) if self._train_dataset else 0
             val_len = len(self._val_dataset) if self._val_dataset else 0
             test_len = len(self._test_dataset) if self._test_dataset else 0
-            s += f"  Datasets (Train/Val/Test): {train_len} / {val_len} / {test_len}\n"
+            s += f"  Datasets (Train|Val|Test): {train_len} | {val_len} | {test_len}\n"
             
         return s
 
@@ -1051,7 +1084,7 @@ class _ObjectDetectionDataset(Dataset):
         self.annotation_paths = annotation_paths
         self.transform = transform
         
-        # --- Propagate 'classes' if they exist (for MLTrainer) ---
+        # --- Propagate 'classes' if they exist ---
         self.classes: List[str] = []
 
     def __len__(self):
@@ -1138,7 +1171,7 @@ class _OD_PairedRandomHorizontalFlip:
         return image, target
 
 
-class ObjectDetectionDatasetMaker(_BaseMaker):
+class DragonDatasetObjectDetection:
     """
     Creates processed PyTorch datasets for object detection from image
     and JSON annotation folders.
@@ -1151,7 +1184,7 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
     so this class provides a `collate_fn` to be used with a DataLoader.
     
     Workflow:
-    1. `maker = ObjectDetectionDatasetMaker.from_folders(img_dir, ann_dir)`
+    1. `maker = DragonDatasetObjectDetection.from_folders(img_dir, ann_dir)`
     2. `maker.set_class_map({'background': 0, 'person': 1, 'car': 2})`
     3. `maker.split_data(val_size=0.2)`
     4. `maker.configure_transforms()`
@@ -1165,7 +1198,9 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
         """
         Typically not called directly. Use the class method `from_folders()` to create an instance.
         """
-        super().__init__()
+        self._train_dataset = None
+        self._test_dataset = None
+        self._val_dataset = None
         self.image_paths: List[Path] = []
         self.annotation_paths: List[Path] = []
         self.class_map: Dict[str, int] = {}
@@ -1178,7 +1213,7 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
         self._has_mean_std: bool = False
 
     @classmethod
-    def from_folders(cls, image_dir: Union[str, Path], annotation_dir: Union[str, Path]) -> 'ObjectDetectionDatasetMaker':
+    def from_folders(cls, image_dir: Union[str, Path], annotation_dir: Union[str, Path]) -> 'DragonDatasetObjectDetection':
         """
         Creates a maker instance by loading all matching image-annotation pairs
         from two corresponding directories.
@@ -1195,7 +1230,7 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
                                          annotation files.
 
         Returns:
-            ObjectDetectionDatasetMaker: A new instance with all pairs loaded.
+            DragonDatasetObjectDetection: A new instance with all pairs loaded.
         """
         maker = cls()
         img_path_obj = make_fullpath(image_dir, enforce="directory")
@@ -1242,12 +1277,12 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
         Logs a report of the types, sizes, and channels of image files
         found in the directory.
         """
-        VisionDatasetMaker.inspect_folder(path)
+        DragonDatasetVision.inspect_folder(path)
 
-    def set_class_map(self, class_map: Dict[str, int]) -> 'ObjectDetectionDatasetMaker':
+    def set_class_map(self, class_map: Dict[str, int]) -> 'DragonDatasetObjectDetection':
         """
         Sets a map of class_name -> pixel_value. This is used by the
-        MLTrainer for clear evaluation reports.
+        trainer for clear evaluation reports.
         
         **Important:** For object detection models, 'background' MUST
         be included as class 0.
@@ -1272,7 +1307,7 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
         return []
 
     def split_data(self, val_size: float = 0.2, test_size: float = 0.0, 
-                   random_state: Optional[int] = 42) -> 'ObjectDetectionDatasetMaker':
+                   random_state: Optional[int] = 42) -> 'DragonDatasetObjectDetection':
         """
         Splits the loaded image-annotation pairs into train, validation, and test sets.
 
@@ -1282,7 +1317,7 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
             random_state (int | None): Seed for reproducible splits.
 
         Returns:
-            ObjectDetectionDatasetMaker: The same instance, now with datasets created.
+            DragonDatasetObjectDetection: The same instance, now with datasets created.
         """
         if self._is_split:
             _LOGGER.warning("Data has already been split.")
@@ -1336,7 +1371,7 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
 
     def configure_transforms(self, 
                              mean: Optional[List[float]] = [0.485, 0.456, 0.406], 
-                             std: Optional[List[float]] = [0.229, 0.224, 0.225]) -> 'ObjectDetectionDatasetMaker':
+                             std: Optional[List[float]] = [0.229, 0.224, 0.225]) -> 'DragonDatasetObjectDetection':
         """
         Configures and applies the image and target transformations.
         
@@ -1351,7 +1386,7 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
             std (List[float] | None): The std dev values for image normalization.
 
         Returns:
-            ObjectDetectionDatasetMaker: The same instance, with transforms applied.
+            DragonDatasetObjectDetection: The same instance, with transforms applied.
         """
         if not self._is_split:
             _LOGGER.error("Transforms must be configured AFTER splitting data. Call .split_data() first.")
@@ -1420,8 +1455,8 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
             raise RuntimeError()
 
         if self._test_dataset:
-            return self._train_dataset, self._val_dataset, self._test_dataset
-        return self._train_dataset, self._val_dataset
+            return self._train_dataset, self._val_dataset, self._test_dataset # type: ignore
+        return self._train_dataset, self._val_dataset # type: ignore
     
     @property
     def collate_fn(self) -> Callable:
@@ -1468,8 +1503,21 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
             )
         
         # Save the file
-        save_recipe(recipe, file_path)
-        
+        _save_recipe(recipe, file_path)
+    
+    def images_per_dataset(self) -> str:
+        """
+        Get the number of images per dataset as a string.
+        """
+        if self._is_split:
+            train_len = len(self._train_dataset) if self._train_dataset else 0
+            val_len = len(self._val_dataset) if self._val_dataset else 0
+            test_len = len(self._test_dataset) if self._test_dataset else 0
+            return f"Train | Validation | Test: {train_len} | {val_len} | {test_len} images"
+        else:
+            _LOGGER.warning("No datasets found.")
+            return "No datasets found"
+
     def __repr__(self) -> str:
         s = f"<{self.__class__.__name__}>:\n"
         s += f"  Total Image-Annotation Pairs: {len(self.image_paths)}\n"
@@ -1483,7 +1531,7 @@ class ObjectDetectionDatasetMaker(_BaseMaker):
             train_len = len(self._train_dataset) if self._train_dataset else 0
             val_len = len(self._val_dataset) if self._val_dataset else 0
             test_len = len(self._test_dataset) if self._test_dataset else 0
-            s += f"  Datasets (Train/Val/Test): {train_len} / {val_len} / {test_len}\n"
+            s += f"  Datasets (Train|Val|Test): {train_len} | {val_len} | {test_len}\n"
             
         return s
 

ml_tools/ML_vision_evaluation.py

@@ -18,7 +18,10 @@ from torchmetrics.detection import MeanAveragePrecision
 from .path_manager import make_fullpath
 from ._logger import _LOGGER
 from ._script_info import _script_info
-from .keys import VisionKeys
+from ._keys import VisionKeys
+from .ML_configuration import (BinarySegmentationMetricsFormat,
+                               MultiClassSegmentationMetricsFormat,
+                               _BaseSegmentationFormat)
 
 
 __all__ = [
@@ -26,12 +29,15 @@ __all__ = [
     "object_detection_metrics"
 ]
 
+DPI_value = 250
+
 
 def segmentation_metrics(
     y_true: np.ndarray,
     y_pred: np.ndarray,
     save_dir: Union[str, Path],
-    class_names: Optional[List[str]] = None
+    class_names: Optional[List[str]] = None,
+    config: Optional[Union[BinarySegmentationMetricsFormat, MultiClassSegmentationMetricsFormat]] = None
 ):
     """
     Calculates and saves pixel-level metrics for segmentation tasks.
@@ -48,9 +54,20 @@ def segmentation_metrics(
         y_pred (np.ndarray): Predicted masks (e.g., shape [N, H, W]).
         save_dir (str | Path): Directory to save the metrics report and plots.
         class_names (List[str] | None): Names of the classes for the report.
+        config (object): Formatting configuration object.
     """
     save_dir_path = make_fullpath(save_dir, make=True, enforce="directory")
     
+    # --- Parse Config or use defaults ---
+    if config is None:
+        format_config = _BaseSegmentationFormat()
+    else:
+        format_config = config
+
+    # --- Set Matplotlib font size ---
+    original_rc_params = plt.rcParams.copy()
+    plt.rcParams.update({'font.size': format_config.font_size})
+    
     # Get all unique class labels present in either true or pred
     labels = np.unique(np.concatenate((np.unique(y_true), np.unique(y_pred)))).astype(int)
     
@@ -110,7 +127,7 @@ def segmentation_metrics(
     report_lines.append(per_class_df.to_string(index=False, float_format="%.4f"))
 
     report_string = "\n".join(report_lines)
-    print(report_string)
+    # print(report_string) # <-- I removed the print(report_string)
     
     # Save text report
     save_filename = VisionKeys.SEGMENTATION_REPORT + ".txt"
@@ -120,11 +137,11 @@ def segmentation_metrics(
 
     # --- 3. Save Per-Class Metrics Heatmap ---
     try:
-        plt.figure(figsize=(max(8, len(labels) * 0.5), 6), dpi=100)
+        plt.figure(figsize=(max(8, len(labels) * 0.5), 6), dpi=DPI_value)
         sns.heatmap(
             per_class_df.set_index('Class').T, 
             annot=True, 
-            cmap='viridis', 
+            cmap=format_config.heatmap_cmap, # Use config cmap
             fmt='.3f',
             linewidths=0.5
         )
@@ -149,7 +166,11 @@ def segmentation_metrics(
             confusion_matrix=cm,
             display_labels=display_names
         )
-        disp.plot(cmap='Blues', ax=ax_cm, xticks_rotation=45)
+        disp.plot(cmap=format_config.cm_cmap, ax=ax_cm, xticks_rotation=45) # Use config cmap
+        
+        # Manually update font size of cell texts
+        for text in disp.text_.flatten(): # type: ignore
+            text.set_fontsize(format_config.font_size)
         
         ax_cm.set_title("Pixel-Level Confusion Matrix")
         plt.tight_layout()
@@ -160,6 +181,9 @@ def segmentation_metrics(
         plt.close(fig_cm)
     except Exception as e:
         _LOGGER.error(f"Could not generate confusion matrix: {e}")
+        
+    # --- Restore RC params ---
+    plt.rcParams.update(original_rc_params)
 
 
 def object_detection_metrics(