✨ Add Anomalib v1 tasks to OTX v2

[ashwinvaidya17]

Summary

Remaining

• Some files can be refactored and moved to separate folder to make them more readable
• Performance numbers are lower than Anomalib. Need to investigate the cause.

How to test

import torch
from torchvision.transforms.v2 import Normalize, ToDtype

from otx.algo.anomaly.padim import Padim
from otx.algo.anomaly.stfpm import STFPM
from otx.core.data.transform_libs.torchvision import PadtoSquare, ResizetoLongestEdge
from otx.core.model.entity.anomaly import OTXAnomalyModel, OVAnomalyModel
from otx.core.types.export import OTXExportFormatType
from otx.core.types.task import OTXTaskType
from otx.data.anomaly import AnomalyDataModule
from otx.engine import Engine

transforms = [
    ResizetoLongestEdge(256, antialias=True),
    PadtoSquare(),
    ToDtype(torch.float32),
    Normalize(mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]),
]
model = Padim(input_size=(256, 256), layers=["layer1", "layer2", "layer3"], backbone="resnet18")
task = OTXTaskType.ANOMALY_CLASSIFICATION
engine = Engine(
        datamodule=datamodule,
        task=task,
        model=model,
        val_check_interval=1.0,
        num_sanity_val_steps=0,
    )
engine.train(max_epochs=1)
engine.test()
# OpenVINO
engine.export(export_format=OTXExportFormatType.OPENVINO)
model = OVAnomalyModel(
  model_name="otx-workspace/exported_model.xml",
  async_inference=True,
  use_throughput_mode=False,
)

engine = Engine(
    model=model,
    datamodule=datamodule,
    task=task,
)
engine.test()

Checklist

• I have added unit tests to cover my changes.​
• I have added integration tests to cover my changes.​
• I have added e2e tests for validation.
• I have added the description of my changes into CHANGELOG in my target branch (e.g., CHANGELOG in develop).​
• I have updated the documentation in my target branch accordingly (e.g., documentation in develop).
• I have linked related issues.

License

• I submit my code changes under the same Apache License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below).

# Copyright (C) 2023 Intel Corporation
# SPDX-License-Identifier: Apache-2.0

[harimkang]

Thanks for doing this, I have a few comments.

[ashwinvaidya17]

I've pushed new changes. Not sure if this helps for now but maybe with the OTXModel refactor it might become better. But it would be nice to get feedback from you all.
I still have to extend changes to Anomaly OpenVINO inference.

[harimkang]

I've pushed new changes. Not sure if this helps for now but maybe with the OTXModel refactor it might become better. But it would be nice to get feedback from you all. I still have to extend changes to Anomaly OpenVINO inference.

If possible, could you please separate IR-related functionality into the next PR?

[ashwinvaidya17]

I've pushed new changes. Not sure if this helps for now but maybe with the OTXModel refactor it might become better. But it would be nice to get feedback from you all. I still have to extend changes to Anomaly OpenVINO inference.

If possible, could you please separate IR-related functionality into the next PR?

Do you want me to remove the export stuff from this PR?

[harimkang]

I've pushed new changes. Not sure if this helps for now but maybe with the OTXModel refactor it might become better. But it would be nice to get feedback from you all. I still have to extend changes to Anomaly OpenVINO inference.

If possible, could you please separate IR-related functionality into the next PR?

Do you want me to remove the export stuff from this PR?

I'm not saying you have to do that, but if you think the IR is going to take a long time. It seems like a good idea to separate things that are going to take a long time. (It doesn't look good to have PRs open for too long.)

If you do separate them, you can skip the IR-related integration tests for now.

[samet-akcay]

@ashwinvaidya17, thanks a lot for the great effort! Much appreciated.

I've added some minor comments, and some questions.

[ashwinvaidya17]

Can we skip python 3.9 checks for anomaly task? Anomalib v1 supports 3.10 and above

[samet-akcay]

Thanks for the effort!

[harimkang]

Can we skip python 3.9 checks for anomaly task? Anomalib v1 supports 3.10 and above

I talked to Mark and Emily about this, and we decided that OTX 2.0 will also not support python 3.9 for now. So I'm going to remove it from the CI test.

[harimkang]

Can we skip python 3.9 checks for anomaly task? Anomalib v1 supports 3.10 and above

i create PR (#2991) - Merged

[harimkang]

Can you please add integration tests to CI in the next PR? I think we need to add tasks to tox.ini and .github/workflows/pre_merge.yaml.