Upgrade to new torch metrics

anomalib/models/components/base/anomaly_module.py

@@ -21,7 +21,7 @@
 from omegaconf import DictConfig, ListConfig
 from pytorch_lightning.callbacks.base import Callback
 from torch import Tensor, nn
-from torchmetrics import F1, MetricCollection
+from torchmetrics import F1Score, MetricCollection
 
 from anomalib.utils.metrics import (
     AUROC,
@@ -59,11 +59,11 @@ def __init__(self, params: Union[DictConfig, ListConfig]):
 
         # metrics
         image_auroc = AUROC(num_classes=1, pos_label=1, compute_on_step=False)
-        image_f1 = F1(num_classes=1, compute_on_step=False, threshold=self.hparams.model.threshold.image_default)
+        image_f1 = F1Score(num_classes=1, compute_on_step=False, threshold=self.hparams.model.threshold.image_default)
         pixel_auroc = AUROC(num_classes=1, pos_label=1, compute_on_step=False)
-        pixel_f1 = F1(num_classes=1, compute_on_step=False, threshold=self.hparams.model.threshold.pixel_default)
-        self.image_metrics = MetricCollection([image_auroc, image_f1], prefix="image_").cpu()
-        self.pixel_metrics = MetricCollection([pixel_auroc, pixel_f1], prefix="pixel_").cpu()
+        pixel_f1 = F1Score(num_classes=1, compute_on_step=False, threshold=self.hparams.model.threshold.pixel_default)
+        self.image_metrics = MetricCollection([image_auroc, image_f1], prefix="image_", compute_groups=False).cpu()
+        self.pixel_metrics = MetricCollection([pixel_auroc, pixel_f1], prefix="pixel_", compute_groups=False).cpu()
 
     def forward(self, batch):  # pylint: disable=arguments-differ
         """Forward-pass input tensor to the module.
@@ -154,8 +154,8 @@ def _compute_adaptive_threshold(self, outputs):
         else:
             self.pixel_threshold.value = self.image_threshold.value
 
-        self.image_metrics.F1.threshold = self.image_threshold.value.item()
-        self.pixel_metrics.F1.threshold = self.pixel_threshold.value.item()
+        self.image_metrics.F1Score.threshold = self.image_threshold.value.item()
+        self.pixel_metrics.F1Score.threshold = self.pixel_threshold.value.item()
 
     def _collect_outputs(self, image_metric, pixel_metric, outputs):
         for output in outputs:

anomalib/utils/callbacks/cdf_normalization.py

@@ -35,8 +35,8 @@ def __init__(self):
 
     def on_test_start(self, _trainer: pl.Trainer, pl_module: AnomalyModule) -> None:
         """Called when the test begins."""
-        pl_module.image_metrics.F1.threshold = 0.5
-        pl_module.pixel_metrics.F1.threshold = 0.5
+        pl_module.image_metrics.F1Score.threshold = 0.5
+        pl_module.pixel_metrics.F1Score.threshold = 0.5
 
     def on_validation_epoch_start(self, trainer: "pl.Trainer", pl_module: AnomalyModule) -> None:
         """Called when the validation starts after training.

anomalib/utils/callbacks/min_max_normalization.py

@@ -29,8 +29,8 @@ class MinMaxNormalizationCallback(Callback):
 
     def on_test_start(self, _trainer: pl.Trainer, pl_module: AnomalyModule) -> None:
         """Called when the test begins."""
-        pl_module.image_metrics.F1.threshold = 0.5
-        pl_module.pixel_metrics.F1.threshold = 0.5
+        pl_module.image_metrics.F1Score.threshold = 0.5
+        pl_module.pixel_metrics.F1Score.threshold = 0.5
 
     def on_validation_batch_end(
         self,

anomalib/utils/callbacks/visualizer_callback.py

@@ -111,7 +111,7 @@ def on_test_batch_end(
             normalize = False  # anomaly maps are already normalized
         else:
             normalize = True  # raw anomaly maps. Still need to normalize
-        threshold = pl_module.pixel_metrics.F1.threshold
+        threshold = pl_module.pixel_metrics.F1Score.threshold
 
         for i, (filename, image, anomaly_map, pred_score, gt_label) in enumerate(
             zip(

requirements/base.txt

@@ -5,6 +5,7 @@ omegaconf>=2.1.1
 opencv-python>=4.5.3.56
 pandas>=1.1.0
 pytorch-lightning>=1.6.0
+torchmetrics>=0.8.0
 torchvision>=0.9.1
 torchtext>=0.9.1
 wandb==0.12.9

tests/pre_merge/utils/callbacks/normalization_callback/test_normalization_callback.py

@@ -42,7 +42,7 @@ def test_normalizer(path=get_dataset_path(), category="shapes"):
     results_with_minmax_normalization = run_train_test(config)
 
     # performance should be the same
-    for metric in ["image_AUROC", "image_F1"]:
+    for metric in ["image_AUROC", "image_F1Score"]:
         assert round(results_without_normalization[0][metric], 3) == round(results_with_cdf_normalization[0][metric], 3)
         assert round(results_without_normalization[0][metric], 3) == round(
             results_with_minmax_normalization[0][metric], 3

tests/pre_merge/utils/metrics/test_adaptive_threshold.py

@@ -65,5 +65,5 @@ def test_non_adaptive_threshold():
 
     trainer = Trainer(**config.trainer, callbacks=callbacks)
     trainer.fit(model=model, datamodule=datamodule)
-    assert trainer.model.image_metrics.F1.threshold == image_threshold
-    assert trainer.model.pixel_metrics.F1.threshold == pixel_threshold
+    assert trainer.model.image_metrics.F1Score.threshold == image_threshold
+    assert trainer.model.pixel_metrics.F1Score.threshold == pixel_threshold