# Copyright The PyTorch Lightning team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import Optional import torch from torch import Tensor from typing_extensions import Literal from torchmetrics.functional.segmentation.utils import _segmentation_inputs_format from torchmetrics.utilities import rank_zero_warn from torchmetrics.utilities.compute import _safe_divide def _dice_score_validate_args( num_classes: int, include_background: bool, average: Optional[Literal["micro", "macro", "weighted", "none"]] = "micro", input_format: Literal["one-hot", "index", "mixed"] = "one-hot", aggregation_level: Optional[Literal["samplewise", "global"]] = "samplewise", ) -> None: """Validate the arguments of the metric.""" if not isinstance(num_classes, int) or num_classes <= 0: raise ValueError(f"Expected argument `num_classes` must be a positive integer, but got {num_classes}.") if not isinstance(include_background, bool): raise ValueError(f"Expected argument `include_background` must be a boolean, but got {include_background}.") allowed_average = ["micro", "macro", "weighted", "none"] if average is not None and average not in allowed_average: raise ValueError(f"Expected argument `average` to be one of {allowed_average} or None, but got {average}.") if input_format not in ["one-hot", "index", "mixed"]: raise ValueError( f"Expected argument `input_format` to be one of 'one-hot', 'index', 'mixed', but got {input_format}." ) if aggregation_level not in ("samplewise", "global"): raise ValueError( f"Expected argument `aggregation_level` to be one of `samplewise`, `global`, but got {aggregation_level}" ) def _dice_score_update( preds: Tensor, target: Tensor, num_classes: int, include_background: bool, input_format: Literal["one-hot", "index", "mixed"] = "one-hot", ) -> tuple[Tensor, Tensor, Tensor]: """Update the state with the current prediction and target.""" preds, target = _segmentation_inputs_format(preds, target, include_background, num_classes, input_format) reduce_axis = list(range(2, target.ndim)) intersection = torch.sum(preds * target, dim=reduce_axis) target_sum = torch.sum(target, dim=reduce_axis) pred_sum = torch.sum(preds, dim=reduce_axis) numerator = 2 * intersection denominator = pred_sum + target_sum support = target_sum return numerator, denominator, support def _dice_score_compute( numerator: Tensor, denominator: Tensor, average: Optional[Literal["micro", "macro", "weighted", "none"]] = "micro", aggregation_level: Optional[Literal["samplewise", "global"]] = "samplewise", support: Optional[Tensor] = None, ) -> Tensor: """Compute the Dice score from the numerator and denominator.""" if aggregation_level == "global": numerator = torch.sum(numerator, dim=0).unsqueeze(0) denominator = torch.sum(denominator, dim=0).unsqueeze(0) support = torch.sum(support, dim=0) if support is not None else None if average == "micro": numerator = torch.sum(numerator, dim=-1) denominator = torch.sum(denominator, dim=-1) return _safe_divide(numerator, denominator, zero_division="nan") dice = _safe_divide(numerator, denominator, zero_division="nan") if average == "macro": return torch.nanmean(dice, dim=-1) if average == "weighted": if not isinstance(support, torch.Tensor): raise ValueError(f"Expected argument `support` to be a tensor, got: {type(support)}.") weights = _safe_divide(support, torch.sum(support, dim=-1, keepdim=True), zero_division="nan") nan_mask = dice.isnan().all(dim=-1) dice = torch.nansum(dice * weights, dim=-1) dice[nan_mask] = torch.nan return dice if average in ("none", None): return dice raise ValueError(f"Invalid value for `average`: {average}.") def dice_score( preds: Tensor, target: Tensor, num_classes: int, include_background: bool = True, average: Optional[Literal["micro", "macro", "weighted", "none"]] = "macro", input_format: Literal["one-hot", "index", "mixed"] = "one-hot", aggregation_level: Optional[Literal["samplewise", "global"]] = "samplewise", ) -> Tensor: """Compute the Dice score for semantic segmentation. Args: preds: Predictions from model target: Ground truth values num_classes: Number of classes include_background: Whether to include the background class in the computation average: The method to average the dice score. Options are ``"micro"``, ``"macro"``, ``"weighted"``, ``"none"`` or ``None``. This determines how to average the dice score across different classes. input_format: What kind of input the function receives. Choose between ``"one-hot"`` for one-hot encoded tensors, ``"index"`` for index tensors or ``"mixed"`` for one one-hot encoded and one index tensor aggregation_level: The level at which to aggregate the dice score. Options are ``"samplewise"`` or ``"global"``. For ``"samplewise"`` the dice score is computed for each sample and then averaged. For ``"global"`` the dice score is computed globally over all samples. Returns: The Dice score. Example (with one-hot encoded tensors): >>> from torch import randint >>> from torchmetrics.functional.segmentation import dice_score >>> _ = torch.manual_seed(42) >>> preds = randint(0, 2, (4, 5, 16, 16)) # 4 samples, 5 classes, 16x16 prediction >>> target = randint(0, 2, (4, 5, 16, 16)) # 4 samples, 5 classes, 16x16 target >>> # dice score micro averaged over all classes >>> dice_score(preds, target, num_classes=5, average="micro") tensor([0.4842, 0.4968, 0.5053, 0.4902]) >>> # dice score per sample and class >>> dice_score(preds, target, num_classes=5, average="none") tensor([[0.4724, 0.5185, 0.4710, 0.5062, 0.4500], [0.4571, 0.4980, 0.5191, 0.4380, 0.5649], [0.5428, 0.4904, 0.5358, 0.4830, 0.4724], [0.4715, 0.4925, 0.4797, 0.5267, 0.4788]]) >>> # global dice score over all samples with macro averaging >>> dice_score(preds, target, num_classes=5, average="macro", aggregation_level="global") tensor([0.4942]) Example (with index tensors): >>> from torch import randint >>> from torchmetrics.functional.segmentation import dice_score >>> _ = torch.manual_seed(42) >>> preds = randint(0, 5, (4, 16, 16)) # 4 samples, 5 classes, 16x16 prediction >>> target = randint(0, 5, (4, 16, 16)) # 4 samples, 5 classes, 16x16 target >>> # dice score micro averaged over all classes >>> dice_score(preds, target, num_classes=5, average="micro", input_format="index") tensor([0.2031, 0.1914, 0.2266, 0.1641]) >>> # dice score per sample and class >>> dice_score(preds, target, num_classes=5, average="none", input_format="index") tensor([[0.1731, 0.1667, 0.2400, 0.2424, 0.1947], [0.2245, 0.2247, 0.2321, 0.1132, 0.1682], [0.2500, 0.2476, 0.1887, 0.1818, 0.2718], [0.1308, 0.1800, 0.1980, 0.1607, 0.1522]]) >>> # global dice score over all samples with macro averaging >>> dice_score(preds, target, num_classes=5, average="macro", aggregation_level="global", input_format="index") tensor([0.1965]) """ if average == "micro": rank_zero_warn( "dice_score metric currently defaults to `average=micro`, but will change to" "`average=macro` in the v1.9 release." " If you've explicitly set this parameter, you can ignore this warning.", UserWarning, ) _dice_score_validate_args(num_classes, include_background, average, input_format, aggregation_level) numerator, denominator, support = _dice_score_update(preds, target, num_classes, include_background, input_format) return _dice_score_compute(numerator, denominator, average, aggregation_level=aggregation_level, support=support)