ahidalgocenteno/triscore-composer

该数据集是一个多模态嵌入数据集，包含音频、乐谱和图像三种模态的特征表示。具体特征包括：基于MERT模型生成的音频嵌入（audio-embeddings-MERT）、基于RoBERTa模型生成的乐谱嵌入（score-embeddings-RoBERTa）和基于ViT模型生成的图像嵌入（image-embeddings-ViT），每个嵌入均为float32类型的列表。此外，数据集还提供整数类型的标签（label）和字符串类型的标识符（identifier）。数据集分为训练集（1829个样本）、验证集（229个样本）和测试集（229个样本），适用于多模态机器学习任务，如分类或跨模态分析。

This dataset is a multimodal embedding dataset that includes feature representations for three modalities: audio, musical scores, and images. Specific features consist of audio embeddings generated by the MERT model (audio-embeddings-MERT), score embeddings generated by the RoBERTa model (score-embeddings-RoBERTa), and image embeddings generated by the ViT model (image-embeddings-ViT), each represented as a list of float32 values. Additionally, the dataset provides integer labels (label) and string identifiers (identifier). It is divided into a training set (1829 examples), a validation set (229 examples), and a test set (229 examples), suitable for multimodal machine learning tasks such as classification or cross-modal analysis.