fix winclip openvino inference output without mask

[junxnone]

📝 Description

WinClip OpenVINO Inference output is only the classification score.

• 🛠️ Fixes [Task]: Enable WinCLIP export #1641

✨ Changes

Select what type of change your PR is:

• 🐞 Bug fix (non-breaking change which fixes an issue)
• 🔨 Refactor (non-breaking change which refactors the code base)
• 🚀 New feature (non-breaking change which adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
• 📚 Documentation update
• 🔒 Security update

✅ Checklist

Before you submit your pull request, please make sure you have completed the following steps:

• 📋 I have summarized my changes in the CHANGELOG and followed the guidelines for my type of change (skip for minor changes, documentation updates, and test enhancements).
• 📚 I have made the necessary updates to the documentation (if applicable).
• 🧪 I have written tests that support my changes and prove that my fix is effective or my feature works (if applicable).

For more information about code review checklists, see the Code Review Checklist.

[adrianboguszewski]

@junxnone please check if it doesn't conflict with the order from DFM: #1952

[junxnone]

• cannot export the torch model for winclip:

anomalib_winclip/train.py", line 21, in <module>
    export_torch = engine.export(model,ExportType.TORCH)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anomalib_org/src/anomalib/engine/engine.py", line 937, in export
    exported_model_path = model.to_torch(
                          ^^^^^^^^^^^^^^^
  File "/anomalib_org/src/anomalib/models/components/base/export_mixin.py", line 83, in to_torch
    torch.save(
  File "/miniforge3/envs/tanomalib_org/lib/python3.11/site-packages/torch/serialization.py", line 628, in save
    _save(obj, opened_zipfile, pickle_module, pickle_protocol, _disable_byteorder_record)
  File "/miniforge3/envs/tanomalib_org/lib/python3.11/site-packages/torch/serialization.py", line 840, in _save
    pickler.dump(obj)
AttributeError: Can't pickle local object 'WinClipModel.encode_image.<locals>.get_feature_map.<locals>.hook'

• my export code(based on the main branch of anomalib)

from anomalib.data import MVTec
from anomalib.models import WinClip
from anomalib.engine import Engine
from anomalib.deploy import ExportType

datamodule = MVTec()
model = WinClip()
engine = Engine()

engine.fit(datamodule=datamodule, model=model)

export_torch = engine.export(model,ExportType.TORCH)

[junxnone]

root cause for this issue is here:

• only get the first output as output_blob :
  • compiled_model.outputs: [<ConstOutput: names[output] shape[?] type: f32>, <ConstOutput: names[13015] shape[?,240,240] type: f32>]

anomalib/src/anomalib/deploy/inferencers/openvino_inferencer.py

Line 150 in e5b91d6

output_blob = compile_model.output(0)

anomalib/src/anomalib/deploy/inferencers/openvino_inferencer.py

Line 109 in e5b91d6

self.input_blob, self.output_blob, self.model = self.load_model(path)

    11:22:44.293023 line       151         output_blob = compile_model.output(0)
    New var:....... output_blob = <ConstOutput: names[output] shape[?] type: f32>

• if the shape of the first output is 1, will set the task to CLASSIFICATION

anomalib/src/anomalib/deploy/inferencers/openvino_inferencer.py

Lines 257 to 268 in e5b91d6

	predictions = predictions[self.output_blob]
	# Initialize the result variables.
	anomaly_map: np.ndarray | None = None
	pred_label: LabelName | None = None
	pred_mask: float | None = None
	# If predictions returns a single value, this means that the task is
	# classification, and the value is the classification prediction score.
	if len(predictions.shape) == 1:
	task = TaskType.CLASSIFICATION
	pred_score = predictions

    11:22:45.720814 line       258         predictions = predictions[self.output_blob]
    Modified var:.. predictions = array([0.5035419], dtype=float32)
    11:22:45.721448 line       261         anomaly_map: np.ndarray | None = None
    New var:....... anomaly_map = None
    11:22:45.721725 line       262         pred_label: LabelName | None = None
    New var:....... pred_label = None
    11:22:45.721992 line       263         pred_mask: float | None = None
    New var:....... pred_mask = None
    11:22:45.722259 line       267         if len(predictions.shape) == 1:
    11:22:45.722547 line       268             task = TaskType.CLASSIFICATION
    New var:....... task = <TaskType.CLASSIFICATION: 'classification'>

• So, OpenVINO Inferencer will output the pred score only

[junxnone]

Looks like the pred_score for segmentation task is not the correct way to get:

anomalib/src/anomalib/deploy/inferencers/openvino_inferencer.py

Lines 266 to 272 in e5b91d6

	if len(predictions.shape) == 1:
	task = TaskType.CLASSIFICATION
	pred_score = predictions
	else:
	task = TaskType.SEGMENTATION
	anomaly_map = predictions.squeeze()
	pred_score = anomaly_map.reshape(-1).max()