GGT-100K
收藏资源简介：
license: cc-by-nc-nd-4.0 --- <div align="center"> <!-- TODO: Optional logo --> <!-- <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/logo.png" alt="GGT-100K logo" width="72%"> </p> --> <h1 align="center" style="font-size: 32px; font-weight: 900;"> <strong>GGT-100K: Generative Ground Truth for Generalizable Real-World Image Restoration</strong> </h1> <p align="center"><i>Real-world LQ–HQ pairs from MFMs to expand IR generalization boundaries.</i></p> [](https://arxiv.org/abs/2605.31039) [](https://huggingface.co/datasets/VCLab-PolyU/GGT-100K/tree/main) [](https://pan.baidu.com/s/1d-wly8RDoCOi59kBL5reSQ?pwd=f38z) [](https://github.com/PolyU-VCLab/GGT-100K) [](https://polyu-vclab.github.io/GGT-100K/) [Xiangtao Kong](https://scholar.google.com/citations?user=lueNzSgAAAAJ&hl=zh-CN)<sup>1,2,*</sup> | [Jixin Zhao](https://scholar.google.com/citations?user=0Z89rfUAAAAJ)<sup>1,2,*</sup> | [Lingchen Sun](https://scholar.google.com/citations?user=ZCDjTn8AAAAJ&hl=zh-CN)<sup>1,2;</sup> | [Rongyuan Wu](https://scholar.google.com/citations?user=A-U8zE8AAAAJ&hl=zh-CN)<sup>1,2;</sup> | [Lei Zhang](https://www4.comp.polyu.edu.hk/~cslzhang/)<sup>1,2,†</sup> <sup>1</sup> The Hong Kong Polytechnic University <sup>2</sup> OPPO Research Institute <sup>*</sup> Equal contribution. <sup>†</sup> Corresponding author. </div> <a id="news"></a> ## 📰 News - **2026-06-01**: Released the [paper](https://arxiv.org/abs/2605.31039). - **2026-05-28**: Released the [GGT-100K dataset](https://huggingface.co/datasets/VCLab-PolyU/GGT-100K), baseline training code, and checkpoints. --- <!-- Optional teaser image --> <!--  --> <video src="https://huggingface.co/datasets/VCLab-PolyU/GGT-100K/resolve/main/demo_video.mp4" autoplay loop muted playsinline width="100%"></video> <p align="center"><b>Demo.</b> Comparing the LQ-GT pairs from GGT-100K. (You can slide it on the <a href="https://polyu-vclab.github.io/GGT-100K" target="_blank">Project Page</a>).</p> <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/overview.png" alt="GGT-100K overview" width="100%"> </p> <p align="center"><em>Overview of GGT-100K.</em></p> <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/compare1.png" alt="GGT-100K compare1" width="100%"> </p> <p align="center"><em> GGT-100K significantly improves the generalization capability of the models to real-world degradations. </em></p> ## 📌 Quick Links - [📰 News](#news) - [🧰 Download GGT-100K Dataset](#dataset) - [🏗️ Construction Process of GGT-100K (including Restoration Evaluation of SOTA MFMs)](#construction) - [🖼️ Experimental Results](#exps) - [📮 Contact](#contact) - [📚 Citation](#citation) <!-- TODO: Optional figures / qualitative results --> <!-- <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/qualitative.jpg" alt="GGT-100K qualitative results" width="100%"> </p> <p align="center"><em>Qualitative results on GGT-100K.</em></p> --> <a id="dataset"></a> ## 🧰 Download GGT-100K Dataset ### Download links - **[Hugging Face](https://huggingface.co/datasets/VCLab-PolyU/GGT-100K/tree/main)** - **[Baidu Disk](https://pan.baidu.com/s/1d-wly8RDoCOi59kBL5reSQ?pwd=f38z)** (password: `f38z`) ### Expected file structure The download links contain **three parts**: - **`GGT-100K`**: the main paired dataset. - **`existing-dataset`**: external/previous datasets used in our paper. We recommend downloading and using it **together** with GGT-100K for training. - **`pretrained-models`**: pretrained checkpoints for baseline models, including **10 models × 2 settings** (**20 checkpoints** in total): trained on **existing data only** vs. trained on **existing data + GGT-100K**. We provide **three JSONL files** that list paired paths using **relative file paths** (relative to the dataset root), for convenient baseline usage: - **Train (existing data, without GGT-100K)**: `train_existing.jsonl` - **Train (existing data + GGT-100K)**: `train_existing_GGT.jsonl` - **Test (GGT-100K-500)**: `test_GGT_500.jsonl` Each line is a pair: ```json {"gt":"relative/path/to/GT.png","lq":"relative/path/to/LQ.png","prompt":""} ``` **Note**: Among the baseline methods in this project, **only Qwen-Image-Edit (`qwen-image-edit`) uses the `prompt` field**. For other methods, `prompt` can be left empty. When using these lists, you should join the relative paths with your local dataset root directory. ### License This dataset is released under the **Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)** license: [License text](https://creativecommons.org/licenses/by-nc-nd/4.0/) --- <a id="construction"></a> ## 🏗️ Construction Process of GGT-100K <details> <summary><strong>Click to expand construction details</strong></summary> <br> <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/collect.png" alt="GGT-100K construction overview" width="100%" loading="lazy"> </p> <p align="center"><em>GGT-100K is constructed by these four steps.</em></p> ### Restoration evaluation of MFMs We evaluate existing MFMs and report the quantitative results below. <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/table1.png" alt="MFM restoration evaluation table" width="100%" loading="lazy"> </p> </details> --- <a id="exps"></a> ## 🖼️ Experimental Results <details> <summary><strong>Click to expand experimental results</strong></summary> <br> To demonstrate the effectiveness of GGT-100K, we train **10 restoration models** with and without GGT-100K, and report quantitative and visual results. ### Quantitative comparison <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/table2.png" alt="Experimental results table" width="100%" loading="lazy"> </p> ### Visual comparison <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/visual_main.png" alt="Visual comparison (main)" width="100%" loading="lazy"> </p> <p align="center"> <img src="https://raw.githubusercontent.com/PolyU-VCLab/GGT-100K/main/docs/static/images/visual2.png" alt="Visual comparison (more)" width="100%" loading="lazy"> </p> </details> --- ## 📮 Contact If you have any questions, please feel free to contact: `xiangtao.kong@connect.polyu.hk` <a id="citation"></a> ## 📚 Citation ``` @article{kong2026GGT-100K, title={GGT-100K: Generative Ground Truth for Generalizable Real-World Image Restoration}, author={Kong, Xiangtao and Zhao, Jixin and Sun, Lingchen and Wu, Rongyuan and Zhang, Lei}, journal={arXiv preprint arXiv:2605.31039}, year={2026} } ```
