Remove model to model comparison

src/anomalib/data/utils/path.py

@@ -146,15 +146,15 @@ def validate_path(
     path: str | Path,
     base_dir: str | Path | None = None,
     should_exist: bool = True,
-    accepted_extensions: tuple[str, ...] | None = None,
+    extensions: tuple[str, ...] | None = None,
 ) -> Path:
     """Validate the path.
 
     Args:
         path (str | Path): Path to validate.
         base_dir (str | Path): Base directory to restrict file access.
         should_exist (bool): If True, do not raise an exception if the path does not exist.
-        accepted_extensions (tuple[str, ...] | None): Accepted extensions for the path. An exception is raised if the
+        extensions (tuple[str, ...] | None): Accepted extensions for the path. An exception is raised if the
             path does not have one of the accepted extensions. If None, no check is performed. Defaults to None.
 
     Returns:
@@ -221,8 +221,8 @@ def validate_path(
             raise PermissionError(msg)
 
     # Check if the path has one of the accepted extensions
-    if accepted_extensions is not None and path.suffix not in accepted_extensions:
-        msg = f"Path extension is not accepted. Accepted extensions: {accepted_extensions}. Path: {path}"
+    if extensions is not None and path.suffix not in extensions:
+        msg = f"Path extension is not accepted. Accepted extensions: {extensions}. Path: {path}"
         raise ValueError(msg)
 
     return path

src/anomalib/metrics/__init__.py

@@ -11,7 +11,7 @@
 import torchmetrics
 from omegaconf import DictConfig, ListConfig
 
-from . import per_image
+from . import pimo
 from .anomaly_score_distribution import AnomalyScoreDistribution
 from .aupr import AUPR
 from .aupro import AUPRO
@@ -20,7 +20,7 @@
 from .f1_max import F1Max
 from .f1_score import F1Score
 from .min_max import MinMax
-from .per_image import AUPIMO, PIMO
+from .pimo import AUPIMO, PIMO
 from .precision_recall_curve import BinaryPrecisionRecallCurve
 from .pro import PRO
 from .threshold import F1AdaptiveThreshold, ManualThreshold
@@ -37,7 +37,7 @@
     "ManualThreshold",
     "MinMax",
     "PRO",
-    "per_image",
+    "pimo",
     "PIMO",
     "AUPIMO",
 ]

src/anomalib/metrics/per_image/__init__.py → src/anomalib/metrics/pimo/__init__.py

@@ -9,12 +9,6 @@
 
 from .enums import StatsOutliersPolicy, StatsRepeatedPolicy, ThresholdMethod
 from .pimo import AUPIMO, PIMO, AUPIMOResult, PIMOResult
-from .utils import (
-    compare_models_pairwise_ttest_rel,
-    compare_models_pairwise_wilcoxon,
-    format_pairwise_tests_results,
-    per_image_scores_stats,
-)
 
 __all__ = [
     # constants
@@ -27,9 +21,4 @@
     # torchmetrics interfaces
     "PIMO",
     "AUPIMO",
-    # utils
-    "compare_models_pairwise_ttest_rel",
-    "compare_models_pairwise_wilcoxon",
-    "format_pairwise_tests_results",
-    "per_image_scores_stats",
 ]

src/anomalib/metrics/per_image/_validate.py → src/anomalib/metrics/pimo/_validate.py

@@ -12,17 +12,12 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import logging
-from collections import OrderedDict
-from typing import TYPE_CHECKING
 
 import torch
 from torch import Tensor
 
 from .utils import images_classes_from_masks
 
-if TYPE_CHECKING:
-    from .pimo import AUPIMOResult
-
 logger = logging.getLogger(__name__)
 
 
@@ -139,10 +134,6 @@ def is_thresh_bounds(thresh_bounds: tuple[float, float]) -> None:
 
 
 def is_anomaly_maps(anomaly_maps: Tensor) -> None:
-    if not isinstance(anomaly_maps, Tensor):
-        msg = f"Expected anomaly maps to be an Tensor, but got {type(anomaly_maps)}"
-        raise TypeError(msg)
-
     if anomaly_maps.ndim != 3:
         msg = f"Expected anomaly maps have 3 dimensions (N, H, W), but got {anomaly_maps.ndim} dimensions"
         raise ValueError(msg)
@@ -156,10 +147,6 @@ def is_anomaly_maps(anomaly_maps: Tensor) -> None:
 
 
 def is_masks(masks: Tensor) -> None:
-    if not isinstance(masks, Tensor):
-        msg = f"Expected masks to be an Tensor, but got {type(masks)}"
-        raise TypeError(msg)
-
     if masks.ndim != 3:
         msg = f"Expected masks have 3 dimensions (N, H, W), but got {masks.ndim} dimensions"
         raise ValueError(msg)
@@ -185,10 +172,6 @@ def is_masks(masks: Tensor) -> None:
 
 
 def is_binclf_curves(binclf_curves: Tensor, valid_threshs: Tensor | None) -> None:
-    if not isinstance(binclf_curves, Tensor):
-        msg = f"Expected binclf curves to be an Tensor, but got {type(binclf_curves)}"
-        raise TypeError(msg)
-
     if binclf_curves.ndim != 4:
         msg = f"Expected binclf curves to be 4D, but got {binclf_curves.ndim}D"
         raise ValueError(msg)
@@ -229,10 +212,6 @@ def is_binclf_curves(binclf_curves: Tensor, valid_threshs: Tensor | None) -> Non
 
 
 def is_images_classes(images_classes: Tensor) -> None:
-    if not isinstance(images_classes, Tensor):
-        msg = f"Expected image classes to be an Tensor, but got {type(images_classes)}."
-        raise TypeError(msg)
-
     if images_classes.ndim != 1:
         msg = f"Expected image classes to be 1D, but got {images_classes.ndim}D."
         raise ValueError(msg)
@@ -258,10 +237,6 @@ def is_images_classes(images_classes: Tensor) -> None:
 
 
 def is_rates(rates: Tensor, nan_allowed: bool) -> None:
-    if not isinstance(rates, Tensor):
-        msg = f"Expected rates to be an Tensor, but got {type(rates)}."
-        raise TypeError(msg)
-
     if rates.ndim != 1:
         msg = f"Expected rates to be 1D, but got {rates.ndim}D."
         raise ValueError(msg)
@@ -307,10 +282,6 @@ def is_rate_curve(rate_curve: Tensor, nan_allowed: bool, decreasing: bool) -> No
 
 
 def is_per_image_rate_curves(rate_curves: Tensor, nan_allowed: bool, decreasing: bool | None) -> None:
-    if not isinstance(rate_curves, Tensor):
-        msg = f"Expected per-image rate curves to be an Tensor, but got {type(rate_curves)}."
-        raise TypeError(msg)
-
     if rate_curves.ndim != 2:
         msg = f"Expected per-image rate curves to be 2D, but got {rate_curves.ndim}D."
         raise ValueError(msg)
@@ -454,177 +425,3 @@ def is_image_class(image_class: int) -> None:
     if image_class not in {0, 1}:
         msg = f"Expected image class to be either 0 for 'normal' or 1 for 'anomalous', but got {image_class}."
         raise ValueError(msg)
-
-
-def is_models_ordered(models_ordered: tuple[str, ...]) -> None:
-    if not isinstance(models_ordered, tuple):
-        msg = f"Expected models ordered to be a tuple, but got {type(models_ordered)}."
-        raise TypeError(msg)
-
-    if len(models_ordered) < 2:
-        msg = f"Expected models ordered to have at least 2 models, but got {len(models_ordered)}."
-        raise ValueError(msg)
-
-    for model_name in models_ordered:
-        if not isinstance(model_name, str):
-            msg = f"Expected model name to be a string, but got {type(model_name)} for model {model_name}."
-            raise TypeError(msg)
-
-        if model_name == "":
-            msg = "Expected model name to be non-empty, but got empty string."
-            raise ValueError(msg)
-
-    num_redundant_models = len(models_ordered) - len(set(models_ordered))
-    if num_redundant_models > 0:
-        msg = f"Expected models ordered to have unique models, but got {num_redundant_models} redundant models."
-        raise ValueError(msg)
-
-
-def is_confidences(confidences: dict[tuple[str, str], float]) -> None:
-    if not isinstance(confidences, dict):
-        msg = f"Expected confidences to be a dict, but got {type(confidences)}."
-        raise TypeError(msg)
-
-    for (model1, model2), confidence in confidences.items():
-        if not isinstance(model1, str):
-            msg = f"Expected model name to be a string, but got {type(model1)} for model {model1}."
-            raise TypeError(msg)
-
-        if not isinstance(model2, str):
-            msg = f"Expected model name to be a string, but got {type(model2)} for model {model2}."
-            raise TypeError(msg)
-
-        if not isinstance(confidence, float):
-            msg = f"Expected confidence to be a float, but got {type(confidence)} for models {model1} and {model2}."
-            raise TypeError(msg)
-
-        if not (0 <= confidence <= 1):
-            msg = f"Expected confidence to be between 0 and 1, but got {confidence} for models {model1} and {model2}."
-            raise ValueError(msg)
-
-
-def joint_validate_models_ordered_and_confidences(
-    models_ordered: tuple[str, ...],
-    confidences: dict[tuple[str, str], float],
-) -> None:
-    num_models = len(models_ordered)
-    expected_num_pairs = num_models * (num_models - 1)
-
-    if len(confidences) != expected_num_pairs:
-        msg = f"Expected {expected_num_pairs} pairs of models, but got {len(confidences)} pairs of models."
-        raise ValueError(msg)
-
-    models_in_confidences = {model for pair_models in confidences for model in pair_models}
-
-    diff = set(models_ordered).symmetric_difference(models_in_confidences)
-    if len(diff) > 0:
-        msg = (
-            "Expected models in confidences to be the same as models ordered, but got models missing in one"
-            f"of them: {diff}."
-        )
-        raise ValueError(msg)
-
-
-def is_scores_per_model_tensor(scores_per_model: dict[str, Tensor] | OrderedDict[str, Tensor]) -> None:
-    first_key_value = None
-
-    for model_name, scores in scores_per_model.items():
-        if scores.ndim != 1:
-            msg = f"Expected scores to be 1D, but got {scores.ndim}D for model {model_name}."
-            raise ValueError(msg)
-
-        num_valid_scores = scores[~torch.isnan(scores)].numel()
-
-        if num_valid_scores < 1:
-            msg = f"Expected at least 1 non-nan score, but got {num_valid_scores} for model {model_name}."
-            raise ValueError(msg)
-
-        if first_key_value is None:
-            first_key_value = (model_name, scores)
-            continue
-
-        first_model_name, first_scores = first_key_value
-
-        # same shape
-        if scores.shape[0] != first_scores.shape[0]:
-            msg = (
-                "Expected scores to have the same number of scores, "
-                f"but got ({model_name}) {scores.shape[0]} != {first_scores.shape[0]} ({first_model_name})."
-            )
-            raise ValueError(msg)
-
-        # `nan` at the same indices
-        if (torch.isnan(scores) != torch.isnan(first_scores)).any():
-            msg = (
-                "Expected `nan` values, if any, to be at the same indices, "
-                f"but there are differences between models {model_name} and {first_model_name}."
-            )
-            raise ValueError(msg)
-
-
-def is_scores_per_model_aupimoresult(
-    scores_per_model: dict[str, "AUPIMOResult"] | OrderedDict[str, "AUPIMOResult"],
-) -> None:
-    first_key_value = None
-
-    for model_name, aupimoresult in scores_per_model.items():
-        if first_key_value is None:
-            first_key_value = (model_name, aupimoresult)
-            continue
-
-        first_model_name, first_aupimoresult = first_key_value
-
-        if aupimoresult.fpr_bounds != first_aupimoresult.fpr_bounds:
-            msg = (
-                "Expected AUPIMOResult objects in scores per model to have the same FPR bounds, "
-                f"but got ({model_name}) {aupimoresult.fpr_bounds} != "
-                f"{first_aupimoresult.fpr_bounds} ({first_model_name})."
-            )
-            raise ValueError(msg)
-
-
-def is_scores_per_model(
-    scores_per_model: dict[str, Tensor]
-    | OrderedDict[str, Tensor]
-    | dict[str, "AUPIMOResult"]
-    | OrderedDict[str, "AUPIMOResult"],
-) -> None:
-    # it has to be imported here to avoid circular imports
-    from .pimo import AUPIMOResult
-
-    if not isinstance(scores_per_model, dict | OrderedDict):
-        msg = f"Expected scores per model to be a dictionary or ordered dictionary, but got {type(scores_per_model)}."
-        raise TypeError(msg)
-
-    if len(scores_per_model) < 2:
-        msg = f"Expected scores per model to have at least 2 models, but got {len(scores_per_model)}."
-        raise ValueError(msg)
-
-    if not all(isinstance(model_name, str) for model_name in scores_per_model):
-        msg = "Expected scores per model to have model names (strings) as keys."
-        raise TypeError(msg)
-
-    first_instance = next(iter(scores_per_model.values()))
-
-    if (
-        isinstance(first_instance, Tensor)
-        and any(not isinstance(scores, Tensor) for scores in scores_per_model.values())
-    ) or (
-        isinstance(first_instance, AUPIMOResult)
-        and any(not isinstance(scores, AUPIMOResult) for scores in scores_per_model.values())
-    ):
-        msg = (
-            "Values in the scores per model dict must have the same type for values (Tensor or AUPIMOResult), "
-            "but more than one type was found."
-        )
-        raise TypeError(msg)
-
-    if isinstance(first_instance, Tensor):
-        is_scores_per_model_tensor(scores_per_model)
-        return
-
-    is_scores_per_model_tensor(
-        {model_name: scores.aupimos for model_name, scores in scores_per_model.items()},
-    )
-
-    is_scores_per_model_aupimoresult(scores_per_model)