GiantMIDI-Piano

该数据集名为GiantMIDI-Piano，包含了来自2,786位作曲家的10,854首钢琴演奏作品，这些作品是从现场录音转录而来，并以MIDI格式编码。该数据集被划分为训练集、验证集和测试集，划分比例为90/10/0。为了增强训练过程，数据集中还应用了时间膨胀、速度移位和转调等数据增强技术。规模上，该数据集共有10,854个表演作品，其任务重点在于钢琴表演的修复和生成。

This dataset is named GiantMIDI-Piano. It contains 10,854 piano performance works from 2,786 composers, which are transcribed from live recordings and encoded in MIDI format. The dataset is split into training, validation and test sets with a 90/10/0 split ratio. To enhance the training process, data augmentation techniques such as time dilation, tempo shifting and key transposition are applied to the dataset. In terms of scale, this dataset has a total of 10,854 performance works, and its core tasks focus on piano performance restoration and generation.