❓ <title>Model conversion reasoning

[hz1012hz]

Your Question

How to convert the patchcore model to onnx and perform inference?

Forum Check

• I have searched the discussions forum for an answer to my question.

[github-actions[bot]]

Thanks for submitting this issue! It has been added to our triage queue. A maintainer will review it shortly.

[rajeshgangireddy]

Hi @hz1012hz
You can use the engine.export() to export models to ONNX.
For inference you can use the Anomalib's OpenVINOInferencer itself (shown below) for inference or ONNX native API ( https://onnx.ai/onnx/repo-docs/PythonAPIOverview.html )

"""MWE script to export models to ONNX and run inference."""

# Copyright (C) 2025 Intel Corporation
# SPDX-License-Identifier: Apache-2.0


import torch
from anomalib.data import MVTecAD
from anomalib.engine import Engine
from anomalib.models import Patchcore
from anomalib.deploy import ExportType, OpenVINOInferencer

model = Patchcore()
engine = Engine()
datamodule = MVTecAD(category="transistor")

# Train the model
engine.fit(datamodule=datamodule, model=model)
results = engine.test(datamodule=datamodule, model=model)

# export to onnx
export_path = engine.export(model=model, export_type=ExportType.ONNX)
print(f"Model exported to: {export_path}")

# OpenVINOInferencer can be used for ONNX models as well
inferencer = OpenVINOInferencer(path=export_path, device="CPU")
# dummy image 
dummy_image = torch.randn(1, 3, 256, 256)
predictions = inferencer.predict(image=dummy_image)
print(f"Predictions Keys: {predictions.keys()}")
# Predictions Keys: ['image', 'gt_label', 'gt_mask', 'mask_path', 'anomaly_map', 'pred_score', 'pred_mask', 'pred_label', 'explanation', 'image_path']