Colorist
Copyright (c) 2026 Sebastian Doerrich, xAILab Bamberg, University of Bamberg

This product is released under the MIT License (see LICENSE). That license covers
the `colorist` package, the `experiments/` harness, `scripts/`, and `examples/`.

It does NOT cover the third-party work listed below. `experiments/reference_methods/`
contains the thirteen comparison methods evaluated in Table 2 of the paper. Each
lives in its own directory; where the upstream project publishes a license, a
verbatim copy of it sits in that directory and governs that directory's contents.

Our files in those directories (`method.py`, and `net.py` where present) are the
harness wrappers and network definitions written at xAILab Bamberg to run each
method under one common interface. They are derived from, and in several cases
closely follow, the upstream implementations cited below. Attribution is given
here rather than claimed away: if you reuse any of these directories, honour the
upstream terms, not ours.

--------------------------------------------------------------------------------
1. Methods whose upstream project publishes a license
--------------------------------------------------------------------------------

AdaIN                     experiments/reference_methods/style_transfer/training_required/adain/
  Huang & Belongie, "Arbitrary Style Transfer in Real-time with Adaptive
  Instance Normalization", ICCV 2017.
  Upstream: https://github.com/naoto0804/pytorch-AdaIN
  License:  MIT, Copyright (c) 2018 Naoto Inoue                    [LICENSE present]

EFDM                      experiments/reference_methods/style_transfer/training_required/efdm/
  Zhang et al., "Exact Feature Distribution Matching for Arbitrary Style
  Transfer and Domain Generalization", CVPR 2022.
  Upstream: https://github.com/YBZh/EFDM
  License:  MIT, Copyright (c) 2022 Yabin Zhang                    [LICENSE present]

IEContrAST                experiments/reference_methods/style_transfer/training_required/iecontrast/
  Chen et al., "Artistic Style Transfer with Internal-external Learning and
  Contrastive Learning", NeurIPS 2021.
  Upstream: https://github.com/HalbertCH/IEContraAST
  License:  MIT, Copyright (c) 2021 HalbertCH                      [LICENSE present]

SANet                     experiments/reference_methods/style_transfer/training_required/sanet/
  Park & Lee, "Arbitrary Style Transfer with Style-Attentional Networks",
  CVPR 2019.
  Upstream: https://github.com/GlebSBrykin/SANET
  License:  MIT, Copyright (c) 2019 Gleb Brykin                    [LICENSE present]

WCT^2                     experiments/reference_methods/photorealistic_color_transfer/training_free/wct2/
  Yoo et al., "Photorealistic Style Transfer via Wavelet Transforms",
  ICCV 2019.
  Upstream: https://github.com/clovaai/WCT2
  License:  MIT, Copyright (c) 2019 NAVER Corp.                    [LICENSE present]

SGViTs                    experiments/reference_methods/medical_color_transfer/sgvits/
  Doerrich et al., "Self-supervised Vision Transformers are Scalable Generative
  Models for Domain Generalization", MICCAI 2024.
  Upstream: https://github.com/sdoerrich97/vits-are-generative-models
  License:  Apache-2.0                                             [LICENSE present]
  Note: the upstream ViT blocks in turn derive from facebookresearch/deit, /mae
  and /moco-v3, which are cited in the source headers and carry their own terms.

Stylizing ViT             experiments/reference_methods/medical_color_transfer/stylizing_vit/
  Doerrich et al., "Stylizing ViT: Anatomy-Preserving Instance Style Transfer
  for Domain Generalization", ISBI 2026.
  Upstream: https://github.com/sdoerrich97/stylizing-vit
  License:  Apache-2.0                                             [LICENSE present]

--------------------------------------------------------------------------------
2. Methods whose upstream project publishes NO license
--------------------------------------------------------------------------------

The six projects below distribute their reference implementations publicly but
attach no license file, so no terms are granted and default copyright applies to
the upstream code. There is consequently no license for us to reproduce in those
directories. Our files there were written at xAILab Bamberg from the papers and
the public reference code, and are documented as such in their headers. We claim
no rights over the upstream work, and anyone reusing these directories should
contact the upstream authors for terms.

ArtFlow                   experiments/reference_methods/style_transfer/training_required/artflow/
  An et al., "ArtFlow: Unbiased Image Style Transfer via Reversible Neural
  Flows", CVPR 2021.
  Upstream: https://github.com/pkuanjie/ArtFlow                    [no license file]

MAST                      experiments/reference_methods/style_transfer/training_required/mast/
  Deng et al., "Arbitrary Style Transfer via Multi-Adaptation Network",
  ACM MM 2020.
  Upstream: https://github.com/diyiiyiii/Arbitrary-Style-Transfer-via-Multi-Adaptation-Network
                                                                  [no license file]

Styleformer               experiments/reference_methods/style_transfer/training_required/styleformer/
  Wu et al., "StyleFormer: Real-time Arbitrary Style Transfer via Parametric
  Style Composition", ICCV 2021.
  Upstream: https://github.com/Wxl-stars/PytorchStyleFormer        [no license file]

StyTr^2                   experiments/reference_methods/style_transfer/training_required/stytr2/
  Deng et al., "StyTr^2: Image Style Transfer with Transformers", CVPR 2022.
  Upstream: https://github.com/diyiiyiii/StyTR-2                   [no license file]

ModFlows                  experiments/reference_methods/photorealistic_color_transfer/training_free/modflows/
  Larchenko et al., "Color Transfer with Modulated Flows", AAAI 2025.
  Upstream: https://github.com/maria-larchenko/modflows            [no license file]

ContriMix                 experiments/reference_methods/medical_color_transfer/contrimix/
  Nguyen et al., "ContriMix: Scalable stain color augmentation for domain
  generalization without domain labels in digital pathology", COMPAY 2024.
  Upstream: https://gitlab.com/huutan86/intraminibatch_permutation_drit
                                                                  [no license file]

--------------------------------------------------------------------------------
3. Reimplemented algorithm suites
--------------------------------------------------------------------------------

experiments/reference_methods/domainbed/ and experiments/reference_methods/sdg/
contain domain-generalization algorithms (ERM, RSC, SD, SelfReg, IB-ERM, and the
single-domain-generalization set) reimplemented at xAILab Bamberg against an
interface modelled on DomainBed. They are our code, released under this
repository's MIT license, and are not copies of the DomainBed sources. The
interface they follow originates in:

  Gulrajani & Lopez-Paz, "In Search of Lost Domain Generalization", ICLR 2021.
  https://github.com/facebookresearch/DomainBed  (MIT)

Each algorithm's own paper is cited in the module that implements it.

--------------------------------------------------------------------------------
4. Pretrained weights
--------------------------------------------------------------------------------

No third-party weights are distributed in this repository. Reproducing Table 2
requires each method's own checkpoint, obtained from its upstream project under
that project's terms; `experiments/reference_methods/pretrain.py` trains the
training-required methods from a VGG-19 initialisation instead.

The DenseNet121 classifier checkpoints we publish on HuggingFace are our own and
are covered by this repository's MIT license.
