Initial SmoothStyle

.gitignore

@@ -0,0 +1,6 @@
+.DS_Store
+__pycache__/
+*.py[cod]
+.cache/
+.smoothstyle-build-*/
+

CHANGELOG.md

@@ -0,0 +1,11 @@
+# Changelog
+
+## Unreleased
+
+- Created the standalone Hugging Face repository structure.
+- Preserved the existing 80/20 pair-level train/test split.
+- Removed redundant content/style copies from target groups.
+- Added per-example metadata, per-asset provenance status, split manifests, statistics, and checksums.
+- Added mixed-rights licensing and third-party notices.
+- Recorded the content images as a random subset of OmniStyle-150K and preserved its declared Apache-2.0 license and citation.
+- Added Simplified Chinese versions of the README, dataset rights notice, and third-party notices.

DATASET_LICENSE.md

@@ -0,0 +1,43 @@
+# SmoothStyle Dataset Rights and License Notice
+
+[English](DATASET_LICENSE.md) | [简体中文](DATASET_LICENSE.zh-CN.md)
+
+SmoothStyle contains project-authored metadata, project-generated images, third-party reference images, and images derived from combinations of those inputs. It is therefore not covered by one blanket license.
+
+## Project-authored metadata
+
+To the extent owned by the SmoothStyle contributors, the dataset indices, split manifests, statistics, documentation, and provenance annotations are made available under the Creative Commons Attribution 4.0 International license (CC BY 4.0).
+
+Canonical license URL: https://creativecommons.org/licenses/by/4.0/
+
+## OmniStyle-150K content images
+
+SmoothStyle content images were randomly selected from the `content/` collection of OmniStyle-150K:
+
+https://huggingface.co/datasets/StyleXX/OmniStyle-150k
+
+The OmniStyle-150K Hugging Face Dataset Card declares the dataset under Apache-2.0. SmoothStyle redistributes this subset on the basis of that upstream declaration, preserves attribution to OmniStyle-150K, and grants no additional rights in the upstream material.
+
+The OmniStyle paper describes its content-image collection as generated using FLUX. The upstream Dataset Card does not provide per-image model version, prompt, seed, or separate output-license records. Those limitations are retained in SmoothStyle's provenance notice.
+
+## Project-generated style images
+
+Style reference images whose IDs begin with `1` were generated by the SmoothStyle project. Their final public license remains pending until the model/provider terms, prompts, reference inputs, and contributor permissions have been documented. No rights beyond those held by the SmoothStyle contributors are granted.
+
+## Style30K images
+
+Style reference images whose IDs begin with `0` were selected from Style30K. They are third-party materials. Their inclusion does not transfer copyright to the SmoothStyle contributors, and SmoothStyle grants no additional rights in them.
+
+The StyleTokenizer repository is presented under CC BY 4.0 and states that some Style30K images were sourced via LAION-5B. LAION-5B applies CC BY 4.0 to its metadata, not to the underlying images. Underlying image rights remain with their respective original rightsholders and may be subject to additional source-site terms.
+
+## Stylized targets
+
+Stylized targets were generated by the project using the STROTSS method. To the extent copyright or related rights exist in a target, its rights status may depend on both its content reference and its style reference. Generating a target does not extinguish third-party rights in either input.
+
+No blanket license is granted for a stylized target whose input rights have not been verified.
+
+## No implied rights
+
+Nothing in this notice grants rights in trademarks, publicity rights, privacy rights, moral rights, or other rights held by third parties. A source label or license label is a provenance statement, not a warranty that every possible use is permitted.
+
+Before public release, unresolved assets should either be cleared, removed, or distributed only through a mechanism supported by an appropriate legal basis. Access gating alone does not supply missing redistribution rights.

DATASET_LICENSE.zh-CN.md

@@ -0,0 +1,46 @@
+# SmoothStyle 数据集权利与许可声明
+
+[English](DATASET_LICENSE.md) | [简体中文](DATASET_LICENSE.zh-CN.md)
+
+> 本文件是 [DATASET_LICENSE.md](DATASET_LICENSE.md) 的中文译文，仅供理解和参考。如中文译文与英文原文存在不一致，以英文原文为准。
+
+SmoothStyle 包含项目原创元数据、项目生成图像、第三方参考图，以及由这些输入组合生成的派生图像。因此，本数据集不适用单一、统一的概括性许可证。
+
+## 项目原创元数据
+
+在 SmoothStyle 贡献者拥有相关权利的范围内，数据集索引、切分清单、统计信息、文档和来源标注以“知识共享署名 4.0 国际许可协议”（CC BY 4.0）提供。
+
+许可证规范地址：https://creativecommons.org/licenses/by/4.0/
+
+## OmniStyle-150K 内容图
+
+SmoothStyle 内容图从 OmniStyle-150K 的 `content/` 集合中随机选择：
+
+https://huggingface.co/datasets/StyleXX/OmniStyle-150k
+
+OmniStyle-150K 的 Hugging Face Dataset Card 声明该数据集采用 Apache-2.0 许可证。SmoothStyle 基于该上游声明再分发所选子集，保留对 OmniStyle-150K 的署名，不对上游材料额外授予任何权利。
+
+OmniStyle 论文说明其内容图集合使用 FLUX 生成。上游 Dataset Card 未提供逐图模型版本、提示词、随机种子或独立的生成输出许可证记录。SmoothStyle 在来源声明中保留这些限制。
+
+## 项目生成的风格图
+
+ID 以 `1` 开头的风格参考图由 SmoothStyle 项目生成。在完成生成模型或服务条款、提示词、参考输入和贡献者许可记录之前，这些图像的最终公开许可证仍待确定。本项目仅在 SmoothStyle 贡献者实际拥有权利的范围内授予权利。
+
+## Style30K 图像
+
+ID 以 `0` 开头的风格参考图选自 Style30K，属于第三方材料。收录这些图像不会使其版权转移给 SmoothStyle 贡献者，SmoothStyle 也不对这些图像额外授予任何权利。
+
+StyleTokenizer 仓库标示为 CC BY 4.0，并说明 Style30K 的部分图像经由 LAION-5B 获取。LAION-5B 的 CC BY 4.0 仅适用于其元数据，不适用于底层图片。底层图片权利仍属于各自原始权利人，并可能受来源网站附加条款约束。
+
+## 风格化目标图
+
+风格化目标图由项目使用 STROTSS 方法生成。在目标图存在版权或相关权利的范围内，其权利状态可能同时取决于内容参考图和风格参考图。生成目标图不会消除任一输入中存在的第三方权利。
+
+对于输入权利尚未核实的风格化目标图，本项目不授予统一或概括性的许可证。
+
+## 不构成默示授权
+
+本声明中的任何内容均不授予商标权、公开权、隐私权、精神权利或第三方持有的其他权利。来源标签或许可证标签属于来源说明，并不保证所有可能的使用方式均已获得许可。
+
+在公开发布前，应对尚未解决的资产完成清权、将其移除，或者仅在具备适当法律依据的机制下分发。设置访问门槛本身不能补足缺失的再分发权利。
+

LICENSES/APACHE-2.0.txt

@@ -0,0 +1,9 @@
+Apache License 2.0
+
+SPDX identifier: Apache-2.0
+Canonical license text:
+https://www.apache.org/licenses/LICENSE-2.0
+
+This license reference applies to the OmniStyle-150K content-image subset on
+the basis of the upstream dataset's Hugging Face license declaration. It does
+not change the rights status of Style30K images or other third-party material.

LICENSES/CC-BY-4.0.txt

@@ -0,0 +1,10 @@
+Creative Commons Attribution 4.0 International (CC BY 4.0)
+
+SPDX identifier: CC-BY-4.0
+Canonical license and legal code:
+https://creativecommons.org/licenses/by/4.0/
+
+This license applies only to SmoothStyle materials explicitly identified as
+CC BY 4.0. It does not automatically apply to third-party images or to assets
+whose rights status is marked as unknown or pending.
+

README.md

@@ -0,0 +1,161 @@
+---
+pretty_name: SmoothStyle
+license: other
+task_categories:
+  - image-to-image
+tags:
+  - image
+  - style-transfer
+  - controllable-generation
+  - strotss
+size_categories:
+  - 10K<n<100K
+---
+
+# SmoothStyle
+
+[English](README.md) | [简体中文](README.zh-CN.md)
+
+SmoothStyle is an image-to-image style-transfer dataset with continuous style-strength supervision. Each example contains a content image, a style reference image, a stylized target image, and a scalar style strength.
+
+> **Release status:** the repository structure and data indices are prepared for reproducibility, but the dataset must not be made public until the remaining Style30K per-image rights review and project-generated style-image generation records have been completed. See [DATASET_LICENSE.md](DATASET_LICENSE.md) and [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).
+
+## Dataset structure
+
+The dataset contains 1,812 content-style pairs. Each pair has ten stylized targets corresponding to strengths 0.1 through 1.0.
+
+| Split | Pairs | Examples |
+| --- | ---: | ---: |
+| train | 1,449 | 14,490 |
+| test | 363 | 3,630 |
+| total | 1,812 | 18,120 |
+
+Each JSONL row is one training example:
+
+```json
+{
+  "id": "0000_1011_06",
+  "pair_id": "0000_1011",
+  "content_id": "0000",
+  "style_id": "1011",
+  "strength_id": 6,
+  "strength": 0.6,
+  "content_file_name": "content/0000.jpg",
+  "style_file_name": "style/1011.jpg",
+  "target_file_name": "target/s06/0000_1011_6.jpg",
+  "style_source": "project-generated",
+  "generator": "STROTSS"
+}
+```
+
+The `*_file_name` fields follow the Hugging Face ImageFolder metadata convention and are relative to the split directory containing `metadata.jsonl`.
+
+```text
+data/
+├── train/
+│   ├── metadata.jsonl
+│   ├── content/
+│   ├── style/
+│   └── target/s01 ... target/s10/
+└── test/
+    ├── metadata.jsonl
+    ├── content/
+    ├── style/
+    └── target/s01 ... target/s10/
+```
+
+Strength IDs `1` through `10` map to scalar strengths `0.1` through `1.0`. The source collection also used IDs `0` and `11` for duplicated content and style references; those redundant copies are not included in this repository.
+
+## Loading
+
+After publishing this repository on the Hugging Face Hub:
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("ORG/SmoothStyle")
+sample = dataset["train"][0]
+
+content = sample["content"]
+style = sample["style"]
+target = sample["target"]
+strength = sample["strength"]
+```
+
+For a local checkout:
+
+```python
+dataset = load_dataset("imagefolder", data_dir="data")
+```
+
+## Split policy
+
+The existing project split is preserved using seed 42 and an 80/20 split over complete `(content_id, style_id)` pairs. Pair IDs and content IDs do not overlap between train and test. Some style IDs occur in both splits; therefore this split evaluates unseen content-style pairs, not strictly unseen-style generalization. Exact pair lists are stored in `metadata/split_manifest.json`.
+
+## Image sources
+
+Content images were randomly selected from the `content/` collection of [OmniStyle-150K](https://huggingface.co/datasets/StyleXX/OmniStyle-150k). Its Hugging Face Dataset Card declares the dataset under Apache-2.0. The OmniStyle paper describes its content images as generated using FLUX. SmoothStyle preserves this upstream declaration and grants no additional rights in the OmniStyle source material.
+
+Style IDs beginning with `0` were selected from Style30K. Style IDs beginning with `1` were generated by the SmoothStyle project. The complete per-asset status is recorded in `metadata/sources.jsonl`.
+
+According to the StyleTokenizer documentation, some Style30K images were sourced via LAION-5B. LAION-5B licenses its metadata under CC BY 4.0, while the underlying images remain subject to the rights of their original rightsholders. SmoothStyle does not infer that a specific Style30K image came from LAION-5B unless a reliable source mapping is available.
+
+## Generation method
+
+Stylized target images were generated by the dataset authors using an implementation of Style Transfer by Relaxed Optimal Transport and Self-Similarity (STROTSS). Users should cite the STROTSS paper in addition to this dataset.
+
+The target generation operation does not remove rights associated with source content or style images. The rights status of every target therefore depends on the rights status of both source images.
+
+## Intended uses
+
+SmoothStyle is intended for research on controllable image style transfer, continuous style-strength conditioning, image-to-image generation, and related evaluation methods.
+
+The dataset should not be used to claim unseen-style generalization with the provided split, to identify people, or to reproduce or impersonate living artists without appropriate permission.
+
+## Licensing
+
+This is a mixed-rights dataset and is marked `license: other`. Project-authored metadata is offered under CC BY 4.0. No blanket license is granted for third-party images. See:
+
+- [DATASET_LICENSE.md](DATASET_LICENSE.md)
+- [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md)
+- [LICENSES/CC-BY-4.0.txt](LICENSES/CC-BY-4.0.txt)
+- [LICENSES/APACHE-2.0.txt](LICENSES/APACHE-2.0.txt)
+
+## Reproducibility and integrity
+
+- `metadata/dataset_stats.json`: dataset counts and source categories
+- `metadata/split_manifest.json`: deterministic pair membership
+- `metadata/sources.jsonl`: per-asset provenance and rights status
+- `metadata/checksums.sha256`: SHA-256 checksums for generated repository data and metadata
+- `scripts/build_dataset.py`: deterministic build script
+- `scripts/validate_dataset.py`: structural, image-decoding, split, and checksum validation
+
+## Citation
+
+Dataset citation information will be added when the accompanying SmoothStyle paper metadata is finalized.
+
+Please also cite OmniStyle, the source of the content-image subset:
+
+```bibtex
+@inproceedings{wang2025omnistyle,
+  title={OmniStyle: Filtering High Quality Style Transfer Data at Scale},
+  author={Wang, Ye and Liu, Ruiqi and Lin, Jiang and Liu, Fei and Yi, Zili and Wang, Yilin and Ma, Rui},
+  booktitle={CVPR},
+  year={2025}
+}
+```
+
+Please also cite STROTSS:
+
+```bibtex
+@inproceedings{kolkin2019style,
+  title={Style Transfer by Relaxed Optimal Transport and Self-Similarity},
+  author={Kolkin, Nicholas and Salavon, Jason and Shakhnarovich, Greg},
+  booktitle={CVPR},
+  year={2019}
+}
+```
+
+## Contact and takedown
+
+Maintainer contact and a formal correction/takedown address must be added before public release. Until then, this repository should remain private.

README.zh-CN.md

@@ -0,0 +1,148 @@
+# SmoothStyle
+
+[English](README.md) | [简体中文](README.zh-CN.md)
+
+SmoothStyle 是一个具有连续风格强度监督的图像到图像风格迁移数据集。每个样本包含一张内容图、一张风格参考图、一张风格化目标图，以及一个标量形式的风格强度。
+
+> **发布状态：** 当前仓库结构和数据索引已经整理完成，可用于复现；但在完成 Style30K 逐图权利审核以及项目自生成风格图的生成记录之前，不应将该数据集设为公开。详情请参阅 [DATASET_LICENSE.zh-CN.md](DATASET_LICENSE.zh-CN.md) 和 [THIRD_PARTY_NOTICES.zh-CN.md](THIRD_PARTY_NOTICES.zh-CN.md)。
+
+## 数据集结构
+
+数据集包含 1,812 个内容—风格组合。每个组合包含十张风格化目标图，对应从 0.1 到 1.0 的十个风格强度。
+
+| 切分 | 组合数 | 样本数 |
+| --- | ---: | ---: |
+| train | 1,449 | 14,490 |
+| test | 363 | 3,630 |
+| 总计 | 1,812 | 18,120 |
+
+每一行 JSONL 对应一个训练样本：
+
+```json
+{
+  "id": "0000_1011_06",
+  "pair_id": "0000_1011",
+  "content_id": "0000",
+  "style_id": "1011",
+  "strength_id": 6,
+  "strength": 0.6,
+  "content_file_name": "content/0000.jpg",
+  "style_file_name": "style/1011.jpg",
+  "target_file_name": "target/s06/0000_1011_6.jpg",
+  "style_source": "project-generated",
+  "generator": "STROTSS"
+}
+```
+
+`*_file_name` 字段遵循 Hugging Face ImageFolder 的元数据约定，其路径相对于同一切分下包含 `metadata.jsonl` 的目录。
+
+```text
+data/
+├── train/
+│   ├── metadata.jsonl
+│   ├── content/
+│   ├── style/
+│   └── target/s01 ... target/s10/
+└── test/
+    ├── metadata.jsonl
+    ├── content/
+    ├── style/
+    └── target/s01 ... target/s10/
+```
+
+强度 ID `1` 至 `10` 分别映射为标量强度 `0.1` 至 `1.0`。源数据还使用 ID `0` 和 `11` 表示重复的内容参考图和风格参考图；本仓库不包含这些冗余副本。
+
+## 加载方式
+
+将本仓库发布至 Hugging Face Hub 后：
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("ORG/SmoothStyle")
+sample = dataset["train"][0]
+
+content = sample["content"]
+style = sample["style"]
+target = sample["target"]
+strength = sample["strength"]
+```
+
+加载本地仓库：
+
+```python
+dataset = load_dataset("imagefolder", data_dir="data")
+```
+
+## 数据切分策略
+
+本数据集保留项目现有切分：使用随机种子 42，对完整的 `(content_id, style_id)` 组合执行 80/20 切分。train 与 test 之间不存在重复的组合 ID 或内容 ID，但部分风格 ID 同时出现在两个切分中。因此，该切分评估的是未见过的内容—风格组合，而不是严格意义上的未见风格泛化。完整组合列表存放于 `metadata/split_manifest.json`。
+
+## 图像来源
+
+内容图从 [OmniStyle-150K](https://huggingface.co/datasets/StyleXX/OmniStyle-150k) 的 `content/` 集合中随机选择。其 Hugging Face Dataset Card 声明数据集使用 Apache-2.0 许可证。OmniStyle 论文说明这些内容图使用 FLUX 生成。SmoothStyle 保留该上游授权声明，不对 OmniStyle 来源材料额外授予任何权利。
+
+以 `0` 开头的风格 ID 选自 Style30K；以 `1` 开头的风格 ID 由 SmoothStyle 项目生成。每项资产的完整状态记录在 `metadata/sources.jsonl` 中。
+
+根据 StyleTokenizer 文档，Style30K 的部分图像经由 LAION-5B 获取。LAION-5B 以 CC BY 4.0 授权其元数据，但底层图像仍受其原始权利人权利约束。在缺少可靠逐图来源映射时，SmoothStyle 不推断某张具体 Style30K 图像来自 LAION-5B。
+
+## 生成方法
+
+风格化目标图由数据集作者使用“基于松弛最优传输和自相似性的风格迁移”（STROTSS）方法的实现生成。使用本数据集时，除引用本数据集外，还应引用 STROTSS 论文。
+
+目标图生成操作不会消除内容图或风格图所附带的权利。因此，每张目标图的权利状态取决于其内容图和风格图的权利状态。
+
+## 预期用途
+
+SmoothStyle 面向可控图像风格迁移、连续风格强度条件控制、图像到图像生成及相关评估方法的研究。
+
+不应使用当前切分声称模型具备未见风格泛化能力；不得使用本数据集识别个人；未经适当许可，不得利用本数据集复制或冒充在世艺术家的创作。
+
+## 许可说明
+
+本数据集包含多种权利状态，因此 Hugging Face 元数据标记为 `license: other`。项目原创元数据以 CC BY 4.0 提供。本项目不对第三方图像授予统一或概括性的许可证。详情参阅：
+
+- [DATASET_LICENSE.zh-CN.md](DATASET_LICENSE.zh-CN.md)
+- [THIRD_PARTY_NOTICES.zh-CN.md](THIRD_PARTY_NOTICES.zh-CN.md)
+- [LICENSES/CC-BY-4.0.txt](LICENSES/CC-BY-4.0.txt)
+- [LICENSES/APACHE-2.0.txt](LICENSES/APACHE-2.0.txt)
+
+## 可复现性与完整性
+
+- `metadata/dataset_stats.json`：数据集数量统计和来源类别
+- `metadata/split_manifest.json`：确定性的组合切分成员列表
+- `metadata/sources.jsonl`：逐项资产来源和权利状态
+- `metadata/checksums.sha256`：所生成仓库数据和元数据的 SHA-256 校验和
+- `scripts/build_dataset.py`：确定性构建脚本
+- `scripts/validate_dataset.py`：结构、图片解码、切分和校验和验证脚本
+
+## 引用
+
+待 SmoothStyle 配套论文信息确定后，将在此补充本数据集的引用信息。
+
+同时请引用内容图子集的来源 OmniStyle：
+
+```bibtex
+@inproceedings{wang2025omnistyle,
+  title={OmniStyle: Filtering High Quality Style Transfer Data at Scale},
+  author={Wang, Ye and Liu, Ruiqi and Lin, Jiang and Liu, Fei and Yi, Zili and Wang, Yilin and Ma, Rui},
+  booktitle={CVPR},
+  year={2025}
+}
+```
+
+同时请引用 STROTSS：
+
+```bibtex
+@inproceedings{kolkin2019style,
+  title={Style Transfer by Relaxed Optimal Transport and Self-Similarity},
+  author={Kolkin, Nicholas and Salavon, Jason and Shakhnarovich, Greg},
+  booktitle={CVPR},
+  year={2019}
+}
+```
+
+## 联系与下架请求
+
+在公开发布前，必须补充维护者联系方式以及正式的纠错和下架请求邮箱。在完成这些信息之前，本仓库应保持私有。
+

THIRD_PARTY_NOTICES.md

@@ -0,0 +1,47 @@
+# Third-Party Notices
+
+[English](THIRD_PARTY_NOTICES.md) | [简体中文](THIRD_PARTY_NOTICES.zh-CN.md)
+
+## OmniStyle-150K
+
+- Dataset: OmniStyle-150K
+- Repository: https://huggingface.co/datasets/StyleXX/OmniStyle-150k
+- Paper: OmniStyle: Filtering High Quality Style Transfer Data at Scale, CVPR 2025
+- Paper URL: https://openaccess.thecvf.com/content/CVPR2025/html/Wang_OmniStyle_Filtering_High_Quality_Style_Transfer_Data_at_Scale_CVPR_2025_paper.html
+- Declared upstream dataset license: Apache-2.0
+- Use in SmoothStyle: randomly selected content-image subset
+
+The OmniStyle-150K Dataset Card identifies `content/` as its original content-image collection and declares the repository under Apache-2.0. The accompanying paper states that content images were generated using FLUX. SmoothStyle preserves the upstream attribution and license declaration and does not grant additional rights in these images.
+
+The published OmniStyle-150K materials do not provide SmoothStyle with per-image FLUX model-version, prompt, seed, or separate output-license records. SmoothStyle therefore records the license basis as the upstream Dataset Card rather than as an independently verified per-image generation license.
+
+## Style30K / StyleTokenizer
+
+- Project: StyleTokenizer: Defining Image Style by a Single Instance for Controlling Diffusion Models
+- Repository: https://github.com/alipay/style-tokenizer
+- Dataset: Style30K
+- Repository license indicator: CC BY 4.0
+- Use in SmoothStyle: style reference images with IDs beginning with `0`
+- Required action before public release: restore or preserve the mapping to original Style30K filenames and confirm the scope of the dataset redistribution permission
+
+Attribution must identify StyleTokenizer/Style30K, link to its repository and license, and indicate that SmoothStyle uses a selected and renamed subset.
+
+## LAION-5B
+
+- Project: LAION-5B
+- Website: https://laion.ai/blog/laion-5b/
+- Metadata license: CC BY 4.0
+- Underlying images: copyright remains with original rightsholders
+
+The StyleTokenizer documentation states that some Style30K images were sourced via LAION-5B. SmoothStyle does not label an individual image as sourced via LAION-5B without a reliable per-image mapping. Where that mapping is established, the LAION metadata license applies only to the LAION metadata; it does not replace the underlying image license.
+
+## STROTSS
+
+- Paper: Style Transfer by Relaxed Optimal Transport and Self-Similarity
+- Authors: Nicholas Kolkin, Jason Salavon, and Greg Shakhnarovich
+- Paper: https://arxiv.org/abs/1904.12785
+- Use in SmoothStyle: generation of style-strength target images
+
+The STROTSS method citation describes the generation method. It does not determine the copyright status of generated outputs or their source images.
+
+The local project implementation should be reviewed separately before source-code publication because a third-party STROTSS reimplementation may not provide an explicit software license.

THIRD_PARTY_NOTICES.zh-CN.md

@@ -0,0 +1,50 @@
+# 第三方声明
+
+[English](THIRD_PARTY_NOTICES.md) | [简体中文](THIRD_PARTY_NOTICES.zh-CN.md)
+
+> 本文件是 [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md) 的中文译文，仅供理解和参考。如中文译文与英文原文存在不一致，以英文原文为准。
+
+## OmniStyle-150K
+
+- 数据集：OmniStyle-150K
+- 仓库：https://huggingface.co/datasets/StyleXX/OmniStyle-150k
+- 论文：OmniStyle: Filtering High Quality Style Transfer Data at Scale，CVPR 2025
+- 论文地址：https://openaccess.thecvf.com/content/CVPR2025/html/Wang_OmniStyle_Filtering_High_Quality_Style_Transfer_Data_at_Scale_CVPR_2025_paper.html
+- 上游声明的数据集许可证：Apache-2.0
+- 在 SmoothStyle 中的用途：随机选择的内容图子集
+
+OmniStyle-150K Dataset Card 将 `content/` 标识为其原始内容图集合，并声明该仓库采用 Apache-2.0 许可证。配套论文说明内容图使用 FLUX 生成。SmoothStyle 保留上游署名和许可证声明，不对这些图像额外授予权利。
+
+已发布的 OmniStyle-150K 材料没有为 SmoothStyle 提供逐图 FLUX 模型版本、提示词、随机种子或独立输出许可证记录。因此，SmoothStyle 将授权依据记录为上游 Dataset Card，而不是经过独立核实的逐图生成许可证。
+
+## Style30K / StyleTokenizer
+
+- 项目：StyleTokenizer: Defining Image Style by a Single Instance for Controlling Diffusion Models
+- 仓库：https://github.com/alipay/style-tokenizer
+- 数据集：Style30K
+- 仓库许可证标识：CC BY 4.0
+- 在 SmoothStyle 中的用途：ID 以 `0` 开头的风格参考图
+- 公开发布前所需操作：恢复或保留本地 ID 到 Style30K 原始文件名的映射，并确认数据集再分发许可的适用范围
+
+署名信息必须标明 StyleTokenizer/Style30K，提供其仓库和许可证链接，并说明 SmoothStyle 使用的是经过选择和重命名的子集。
+
+## LAION-5B
+
+- 项目：LAION-5B
+- 网站：https://laion.ai/blog/laion-5b/
+- 元数据许可证：CC BY 4.0
+- 底层图像：版权仍属于原始权利人
+
+StyleTokenizer 文档说明 Style30K 的部分图像经由 LAION-5B 获取。若不存在可靠的逐图映射，SmoothStyle 不会将某张具体图片标记为经由 LAION-5B 获取。建立该映射后，LAION 的元数据许可证仍仅适用于 LAION 元数据，不会取代底层图像许可证。
+
+## STROTSS
+
+- 论文：Style Transfer by Relaxed Optimal Transport and Self-Similarity
+- 作者：Nicholas Kolkin、Jason Salavon 和 Greg Shakhnarovich
+- 论文地址：https://arxiv.org/abs/1904.12785
+- 在 SmoothStyle 中的用途：生成具有不同风格强度的目标图
+
+对 STROTSS 方法的引用仅说明数据生成方法，并不决定生成输出或其输入图像的版权状态。
+
+在公开项目源代码前，还应单独审查本地 STROTSS 实现，因为其可能基于未提供明确软件许可证的第三方 STROTSS 重实现。
+