Bump up pytorch-lightning version to 1.6.0 or higher (#193)

* bump up pytorch-lightning version

* bump up to 1.6.0 or higher

* 🛠 Fix typo.

* 🛠 Fix mypy issues.

* 🔄 Switch soft permutation to false by default since this radically
slows down the training.

* 🗑 Remove `self.automatic_optimization=False` from dfkde to automatically set `trainer.global_step`

* 🗑 Remove `self.automatic_optimization=False` from dfm to automatically set `trainer.global_step`

* 🗑 Remove `self.automatic_optimization=False` from padim to automatically set `trainer.global_step`

* 🗑 Remove `self.automatic_optimization=False` from patchcore to automatically set `trainer.global_step`

* 🔄 replace `checkpoint_callback` with `enable_checkpointing`

* ✏️ Edit `check_val_every_n_epoch` from 2 ➡️ 1 to save the weights

* ✏️ Set `check_val_every_n_epoch` to 1 for `fast_run`.

anomalib/models/cflow/config.yaml

@@ -54,9 +54,9 @@ trainer:
   auto_select_gpus: false
   benchmark: false
   check_val_every_n_epoch: 1
-  checkpoint_callback: true
   default_root_dir: null
   deterministic: false
+  enable_checkpointing: true
   fast_dev_run: false
   gpus: 1
   gradient_clip_val: 0

anomalib/models/dfkde/config.yaml

@@ -41,9 +41,9 @@ trainer:
   auto_select_gpus: false
   benchmark: false
   check_val_every_n_epoch: 1 # Don't validate before extracting features.
-  checkpoint_callback: true
   default_root_dir: null
   deterministic: false
+  enable_checkpointing: true
   fast_dev_run: false
   gpus: 1
   gradient_clip_val: 0

anomalib/models/dfkde/model.py

@@ -100,7 +100,6 @@ def __init__(self, hparams: Union[DictConfig, ListConfig]):
             hparams.model.backbone, hparams.model.max_training_points, threshold_steepness, threshold_offset
         )
 
-        self.automatic_optimization = False
         self.embeddings: List[Tensor] = []
 
     @staticmethod

anomalib/models/dfm/config.yaml

@@ -40,9 +40,9 @@ trainer:
   auto_select_gpus: false
   benchmark: false
   check_val_every_n_epoch: 1 # Don't validate before extracting features.
-  checkpoint_callback: true
   default_root_dir: null
   deterministic: false
+  enable_checkpointing: true
   fast_dev_run: false
   gpus: 1
   gradient_clip_val: 0

anomalib/models/dfm/model.py

@@ -34,7 +34,6 @@ def __init__(self, hparams: Union[DictConfig, ListConfig]):
         self.model: DFMModel = DFMModel(
             backbone=hparams.model.backbone, n_comps=hparams.model.pca_level, score_type=hparams.model.score_type
         )
-        self.automatic_optimization = False
         self.embeddings: List[Tensor] = []
 
     @staticmethod

anomalib/models/ganomaly/config.yaml

@@ -61,9 +61,9 @@ trainer:
   auto_select_gpus: false
   benchmark: false
   check_val_every_n_epoch: 2
-  checkpoint_callback: true
   default_root_dir: null
   deterministic: false
+  enable_checkpointing: true
   fast_dev_run: false
   gpus: 1
   gradient_clip_val: 0

anomalib/models/padim/config.yaml

@@ -54,9 +54,9 @@ trainer:
   auto_select_gpus: false
   benchmark: false
   check_val_every_n_epoch: 1 # Don't validate before extracting features.
-  checkpoint_callback: true
   default_root_dir: null
   deterministic: false
+  enable_checkpointing: true
   fast_dev_run: false
   gpus: 1
   gradient_clip_val: 0

anomalib/models/padim/model.py

@@ -294,7 +294,6 @@ def __init__(self, hparams: Union[DictConfig, ListConfig]):
         ).eval()
 
         self.stats: List[Tensor] = []
-        self.automatic_optimization = False
         self.embeddings: List[Tensor] = []
 
     @staticmethod

anomalib/models/patchcore/config.yaml

@@ -52,9 +52,9 @@ trainer:
   auto_select_gpus: false
   benchmark: false
   check_val_every_n_epoch: 1 # Don't validate before extracting features.
-  checkpoint_callback: true
   default_root_dir: null
   deterministic: false
+  enable_checkpointing: true
   fast_dev_run: false
   gpus: 1
   gradient_clip_val: 0

anomalib/models/patchcore/model.py

@@ -275,7 +275,6 @@ def __init__(self, hparams) -> None:
             backbone=hparams.model.backbone,
             apply_tiling=hparams.dataset.tiling.apply,
         )
-        self.automatic_optimization = False
         self.embeddings: List[Tensor] = []
 
     def configure_optimizers(self) -> None:

anomalib/models/stfpm/config.yaml

@@ -61,9 +61,9 @@ trainer:
   auto_select_gpus: false
   benchmark: false
   check_val_every_n_epoch: 1
-  checkpoint_callback: true
   default_root_dir: null
   deterministic: false
+  enable_checkpointing: true
   fast_dev_run: false
   gpus: 1
   gradient_clip_val: 0

anomalib/utils/callbacks/visualizer_callback.py

@@ -28,6 +28,7 @@
 from anomalib.pre_processing.transforms import Denormalize
 from anomalib.utils import loggers
 from anomalib.utils.loggers import AnomalibWandbLogger
+from anomalib.utils.loggers.base import ImageLoggerBase
 
 
 class VisualizerCallback(Callback):
@@ -69,7 +70,7 @@ def _add_images(
         for log_to in module.hparams.project.log_images_to:
             if log_to in loggers.AVAILABLE_LOGGERS:
                 # check if logger object is same as the requested object
-                if log_to in logger_type and module.logger is not None:
+                if log_to in logger_type and module.logger is not None and isinstance(module.logger, ImageLoggerBase):
                     module.logger.add_image(
                         image=visualizer.figure,
                         name=filename.parent.name + "_" + filename.name,

anomalib/utils/loggers/wandb.py

@@ -86,7 +86,7 @@ def __init__(
         anonymous: Optional[bool] = None,
         version: Optional[str] = None,
         project: Optional[str] = None,
-        log_model: Optional[bool] = False,
+        log_model: Union[str, bool] = False,
         experiment=None,
         prefix: Optional[str] = "",
         sync_step: Optional[bool] = None,

requirements/base.txt

@@ -12,7 +12,7 @@ opencv-python>=4.5.3.56
 opencv-contrib-python==4.5.5.62
 pandas~=1.1.5
 pillow==9.0.0
-pytorch-lightning==1.5.9
+pytorch-lightning>=1.6.0
 torch==1.8.1
 torchvision==0.9.1
 torchtext==0.9.1

tests/helpers/model.py

@@ -109,6 +109,7 @@ def setup_model_train(
     # Train the model.
     if fast_run:
         config.trainer.max_epochs = 1
+        config.trainer.check_val_every_n_epoch = 1
 
     trainer = Trainer(callbacks=callbacks, **config.trainer)
     trainer.fit(model=model, datamodule=datamodule)