docs????? GitHub Repository ?????

README.md

@@ -60,7 +60,7 @@ are published separately and were not assigned a retrospective pass/fail label.
 
 The complete method, immutable revisions, and our interpretation of NVIDIA Cosmos
 AnomalyGen are documented in the
-[GitHub repository](https://github.com/kuotunyu/01-defectforge-visa).
+[GitHub repository](https://github.com/kuotunyu/defectforge-visa-synthetic-data).
 
 ## Labels and provenance
 
@@ -117,20 +117,26 @@ unfiltered; they are not relabelled after downstream results.
 
 ## License chain
 
+Source Code 與第三方 Artifact 的完整界線見
+[THIRD_PARTY_NOTICES.md](https://github.com/kuotunyu/defectforge-visa-synthetic-data/blob/main/THIRD_PARTY_NOTICES.md)。
+
 <!-- BEGIN VERIFIED LICENSE_CHAIN -->
-| Asset | License | DefectForge obligation |
+| 資產 | License | DefectForge 義務 |
 |---|---|---|
-| VisA source dataset | CC BY 4.0 | Attribute VisA and its paper; do not include the original images in our HF dataset |
-| `sd2-community/stable-diffusion-2-inpainting` | CreativeML Open RAIL++-M | Preserve the use-based restrictions and disclose the preservation mirror |
-| `diffusers/stable-diffusion-xl-1.0-inpainting-0.1` | CreativeML Open RAIL++-M | Preserve the use-based restrictions |
-| `facebook/dinov2-base` | Apache-2.0 | Attribute the model and DINOv2 paper |
-| DefectForge synthetic images | CC BY 4.0 | Treat as VisA derivatives; retain VisA attribution and disclose diffusion base licenses |
-| DefectForge LoRA weights | CreativeML Open RAIL++-M | Inherit the corresponding base-model restrictions and link the license |
-| DefectForge source code | MIT | The MIT grant applies to code only, not data or model weights |
+| VisA 原始 Dataset | CC BY 4.0 | 標示 VisA 與其論文；Hugging Face Dataset 不得包含原始影像 |
+| `sd2-community/stable-diffusion-2-inpainting` | CreativeML Open RAIL++-M | 保留用途限制，並揭露 preservation mirror |
+| `diffusers/stable-diffusion-xl-1.0-inpainting-0.1` | CreativeML Open RAIL++-M | 保留用途限制 |
+| `facebook/dinov2-base` | Apache-2.0 | 標示模型與 DINOv2 論文 |
+| DefectForge Synthetic Images | CC BY 4.0 | 視為 VisA 衍生內容；保留 VisA attribution，並揭露 Diffusion base model License |
+| DefectForge LoRA Weights | CreativeML Open RAIL++-M | 繼承對應 base model 的限制，並附上 License 連結 |
+| DefectForge Source Code | MIT | MIT 僅授權程式碼，不包含 Dataset 與 Model Weights |
 <!-- END VERIFIED LICENSE_CHAIN -->
 
 ## Citation
 
+DefectForge 的 GitHub Citation Metadata 見
+[CITATION.cff](https://github.com/kuotunyu/defectforge-visa-synthetic-data/blob/main/CITATION.cff)。
+
 - Zou et al., *SPot-the-Difference Self-Supervised Pre-training for Anomaly Detection
   and Segmentation*, ECCV 2022.
 - Oquab et al., *DINOv2: Learning Robust Visual Features without Supervision*, 2023.