README
SVGX-Core-250k Dataset
SVGX-Core-250k is a large-scale vector graphics dataset with detailed annotations.
This dataset includes both simple vector graphics with linear colors and complex vector graphics with gradient fills. For each sample, we provide:
- Structurally optimized SVG files
- Rasterized PNG images (
512x512) - Text captions and emoji descriptions
- Comprehensive metadata
🛠️ How to Use
from datasets import load_dataset
# Login using `huggingface-cli login` if the dataset requires authentication
svgx_dataset = load_dataset("xingxm/SVGX-Core-250k")
📑 Dataset Overview
| Column | Type | Description |
|---|---|---|
blip_caption | string | Caption generated by the BLIP model describing the rendering image. |
qwen_caption | string | Human-like caption generated by the Qwen2.5-VL model. |
image | image | PNG image rendered from the original SVG (512×512 pixels). |
name | string | Short descriptive name (e.g., emoji name). |
source | string | Data source (e.g., "noto-emoji"). |
svg_path | string | Relative file path to the SVG file. |
svg | string | SVG XML content as a string. |
img_analysis | string | Analytical description of the image's visual features. |
uuid | string | Universally unique identifier (UUID) for each sample. |
svg_code | string | Cleaned and normalized SVG content. |
📊 Data Sources
| Source | Approximate Samples |
|---|---|
| Google / Noto-Emoji | ~3,500 |
| Twitter / Color-Emoji | ~3,700 |
| Microsoft / Fluentui-Emoji | ~3,000 |
| Pixso | ~15,000 |
| ReShot | ~30,000 |
| SVGRepo | ~210,000 |
💘 Acknowledgements
SVGX-Core-250k is built upon contributions from:
- Google / Noto-Emoji
- Twitter / Color-Emoji (Twemoji)
- Microsoft / Fluentui-Emoji
- Web-sourced vector graphics from Pixso, ReShot, and SVGRepo
📎 Citation
If you use this code for your research, please cite the following work:
@article{xing2024llm4svg,
title={Empowering LLMs to Understand and Generate Complex Vector Graphics},
author={Xing, Ximing and Hu, Juncheng and Liang, Guotao and Zhang, Jing and Xu, Dong and Yu, Qian},
journal={arXiv preprint arXiv:2412.11102},
year={2024}
}
@article{xing2024svgfusion,
title={SVGFusion: Scalable Text-to-SVG Generation via Vector Space Diffusion},
author={Xing, Ximing and Hu, Juncheng and Zhang, Jing and Xu, Dong and Yu, Qian},
journal={arXiv preprint arXiv:2412.10437},
year={2024}
}
📜 License
This dataset is licensed under the * *Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) ** license.
- ✅ You are free to share and adapt the material.
- ❌ Commercial use is prohibited.
- 📎 Attribution is required.
For full legal terms, please refer to this link.
Contributors
5 commits
Similar repos
Linked in READMEs
README
SVGX-Core-250k Dataset
SVGX-Core-250k is a large-scale vector graphics dataset with detailed annotations.
This dataset includes both simple vector graphics with linear colors and complex vector graphics with gradient fills. For each sample, we provide:
- Structurally optimized SVG files
- Rasterized PNG images (
512x512) - Text captions and emoji descriptions
- Comprehensive metadata
🛠️ How to Use
from datasets import load_dataset
# Login using `huggingface-cli login` if the dataset requires authentication
svgx_dataset = load_dataset("xingxm/SVGX-Core-250k")
📑 Dataset Overview
| Column | Type | Description |
|---|---|---|
blip_caption | string | Caption generated by the BLIP model describing the rendering image. |
qwen_caption | string | Human-like caption generated by the Qwen2.5-VL model. |
image | image | PNG image rendered from the original SVG (512×512 pixels). |
name | string | Short descriptive name (e.g., emoji name). |
source | string | Data source (e.g., "noto-emoji"). |
svg_path | string | Relative file path to the SVG file. |
svg | string | SVG XML content as a string. |
img_analysis | string | Analytical description of the image's visual features. |
uuid | string | Universally unique identifier (UUID) for each sample. |
svg_code | string | Cleaned and normalized SVG content. |
📊 Data Sources
| Source | Approximate Samples |
|---|---|
| Google / Noto-Emoji | ~3,500 |
| Twitter / Color-Emoji | ~3,700 |
| Microsoft / Fluentui-Emoji | ~3,000 |
| Pixso | ~15,000 |
| ReShot | ~30,000 |
| SVGRepo | ~210,000 |
💘 Acknowledgements
SVGX-Core-250k is built upon contributions from:
- Google / Noto-Emoji
- Twitter / Color-Emoji (Twemoji)
- Microsoft / Fluentui-Emoji
- Web-sourced vector graphics from Pixso, ReShot, and SVGRepo
📎 Citation
If you use this code for your research, please cite the following work:
@article{xing2024llm4svg,
title={Empowering LLMs to Understand and Generate Complex Vector Graphics},
author={Xing, Ximing and Hu, Juncheng and Liang, Guotao and Zhang, Jing and Xu, Dong and Yu, Qian},
journal={arXiv preprint arXiv:2412.11102},
year={2024}
}
@article{xing2024svgfusion,
title={SVGFusion: Scalable Text-to-SVG Generation via Vector Space Diffusion},
author={Xing, Ximing and Hu, Juncheng and Zhang, Jing and Xu, Dong and Yu, Qian},
journal={arXiv preprint arXiv:2412.10437},
year={2024}
}
📜 License
This dataset is licensed under the * *Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) ** license.
- ✅ You are free to share and adapt the material.
- ❌ Commercial use is prohibited.
- 📎 Attribution is required.
For full legal terms, please refer to this link.
Similar repos
Linked in READMEs
Contributors
5 commits