arsaporta/symile-m3

Symile-M3是一个多语言的(音频, 图像, 文本)样本数据集。该数据集专门设计用于测试模型在三种不同的高维数据类型之间捕捉高阶信息的能力。通过结合多种语言，构建了一个需要文本和音频来预测图像的任务，并且重要的是，单独的文本或音频都不足以完成任务。数据集包含三种语言变体（2种、5种和10种语言）和四种大小变体（大、中、小和超小）。每个样本包含语言代码、音频数据、图像数据、文本、类别名称、类别ID和目标文本等字段。数据集还包括一个translations.json文件，该文件将ImageNet类别名称映射到所有支持的语言。README文件提供了加载和使用数据集的说明，以及如何处理原始数据和数据集的引用信息。

Symile-M3 is a multilingual dataset of (audio, image, text) samples. The dataset is specifically designed to test a models ability to capture higher-order information between three distinct high-dimensional data types. By incorporating multiple languages, the dataset constructs a task where both text and audio are needed to predict the image, and where neither text nor audio alone would suffice. The dataset is available in three language variants (2, 5, and 10 languages) and four size variants (large, medium, small, and extra small). Each sample in the dataset includes fields such as language code, audio data, image data, text, class name, class ID, and target text. The dataset also includes a translations.json file mapping ImageNet class names across all supported languages. The README provides instructions for loading and using the dataset, as well as information on how to work with the raw data and citations for the dataset.