text2fabric

该数据集名为text2fabric，它将自由文本描述与各种面料材料相连接，包含了15,461条自然语言描述，这些描述与45,000张渲染图像相关联，这些图像展示了3,000种不同的面料材料。这些图像是使用Substance Stager渲染器生成的，既包括程序化材料也包括高质量扫描。该数据集使得大型视觉-语言模型能够专精于特定领域，并提升了诸如细粒度材料检索和自动字幕生成等应用的效果。数据集的规模包括45,000张图像和15,461条描述，其任务是对面料外观的自然语言描述。

The dataset, named text2fabric, links free-text descriptions to various fabric materials. It contains 15,461 natural language descriptions associated with 45,000 rendered images that showcase 3,000 distinct fabric materials. These images were generated using the Substance Stager renderer, covering both procedural materials and high-quality scanned fabric samples. This dataset enables large vision-language models to specialize in targeted domains, and improves the performance of applications such as fine-grained material retrieval and automatic caption generation. With a scale of 45,000 images and 15,461 descriptions, the core task of this dataset focuses on natural language descriptions of fabric appearance.