MatSynth

MatSynth专注于为现代人工智能应用提供物理渲染（PBR）材质数据集。它包含超过4000个超高分辨率的材质，在规模、多样性和细节方面表现出色。这些材质数据包括基础色、漫反射、法线、高度、粗糙度、金属度、镜面反射以及不透明度贴图，并提供材质的来源、许可、类别、标签、创建方法等元数据。MatSynth支持图像到图像、无条件图像生成、图像分类和文本到图像等任务，通过标准化数据操作，可以灵活地进行数据筛选、格式转换，方便用户在PyTorch等框架中使用。该数据集的数据来源于多个在线平台，并经过筛选和增强，部分数据采用CC0和CC-BY许可协议。MatSynth is a physically based rendering (PBR) material dataset tailored for modern artificial intelligence applications. It contains over 4000 ultra-high-resolution materials, excelling in scale, diversity and detail. Each material is paired with albedo, diffuse, normal, height, roughness, metallic, specular reflection and opacity maps, along with metadata including material source, license, category, tags and creation methods. MatSynth supports multiple tasks such as image-to-image translation, unconditional image generation, image classification and text-to-image generation. Through standardized data operations, it allows flexible data filtering and format conversion, making it convenient for users to employ in frameworks like PyTorch. The dataset's data originates from multiple online platforms, and has been screened and enhanced. Part of the data is licensed under CC0 and CC-BY.

搜集汇总

背景与挑战

背景概述

MatSynth是一个针对现代AI应用设计的高质量PBR材质数据集，包含超过4000个超高分辨率材质，提供基础色、法线、粗糙度等多种贴图及丰富的元数据。该数据集支持图像到图像、无条件图像生成、图像分类和文本到图像等多种任务，数据来源于多个在线平台，采用CC0和CC-BY许可协议，方便用户进行灵活的数据筛选和格式转换。