Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix/efficient ad normalize before every validation #1441

Merged
merged 4 commits into from
Nov 5, 2023
Merged

Fix/efficient ad normalize before every validation #1441

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
71cf888
Normalize anomaly maps before every validation
abc-125 Oct 25, 2023
4bfc187
Remove print statement
abc-125 Oct 25, 2023
b0bf342
Merge branch 'main' into fix/efficient-ad-normalize-before-every-vali…
abc-125 Oct 25, 2023
f64be02
Merge branch 'main' into fix/efficient-ad-normalize-before-every-vali…
samet-akcay Nov 3, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 3 additions & 4 deletions src/anomalib/models/efficient_ad/lightning_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,7 +178,7 @@ def map_norm_quantiles(self, dataloader: DataLoader) -> dict[str, Tensor]:
for batch in tqdm.tqdm(dataloader, desc="Calculate Validation Dataset Quantiles", position=0, leave=True):
for img, label in zip(batch["image"], batch["label"]):
if label == 0: # only use good images of validation set!
output = self.model(img.to(self.device))
output = self.model(img.to(self.device), normalize=False)
map_st = output["map_st"]
map_ae = output["map_ae"]
maps_st.append(map_st)
Expand Down Expand Up @@ -256,9 +256,8 @@ def on_validation_start(self) -> None:
"""
Calculate the feature map quantiles of the validation dataset and push to the model.
"""
if (self.current_epoch + 1) == self.trainer.max_epochs:
map_norm_quantiles = self.map_norm_quantiles(self.trainer.datamodule.val_dataloader())
self.model.quantiles.update(map_norm_quantiles)
map_norm_quantiles = self.map_norm_quantiles(self.trainer.datamodule.val_dataloader())
self.model.quantiles.update(map_norm_quantiles)

def validation_step(self, batch: dict[str, str | Tensor], *args, **kwargs) -> STEP_OUTPUT:
"""Validation Step of EfficientAd returns anomaly maps for the input image batch
Expand Down
5 changes: 3 additions & 2 deletions src/anomalib/models/efficient_ad/torch_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -287,10 +287,11 @@ def choose_random_aug_image(self, image: Tensor) -> Tensor:
transform_function = random.choice(transform_functions) # nosec: B311
return transform_function(image, coefficient)

def forward(self, batch: Tensor, batch_imagenet: Tensor = None) -> Tensor | dict:
def forward(self, batch: Tensor, batch_imagenet: Tensor = None, normalize: bool = True) -> Tensor | dict:
"""Prediction by EfficientAd models.

Args:
normalize: Normalize anomaly maps or not.
batch (Tensor): Input images.

Returns:
Expand Down Expand Up @@ -346,7 +347,7 @@ def forward(self, batch: Tensor, batch_imagenet: Tensor = None) -> Tensor | dict
map_st = F.interpolate(map_st, size=(self.input_size[0], self.input_size[1]), mode="bilinear")
map_stae = F.interpolate(map_stae, size=(self.input_size[0], self.input_size[1]), mode="bilinear")

if self.is_set(self.quantiles):
if self.is_set(self.quantiles) and normalize:
map_st = 0.1 * (map_st - self.quantiles["qa_st"]) / (self.quantiles["qb_st"] - self.quantiles["qa_st"])
map_stae = (
0.1 * (map_stae - self.quantiles["qa_ae"]) / (self.quantiles["qb_ae"] - self.quantiles["qa_ae"])
Expand Down
Loading