Syncred-Bench
收藏资源简介:
Syncred-Bench是一个图像数据集,包含两个子集配置:syncred_600(包含600行样本)和fp_450(包含450行样本)。数据以Parquet格式存储,图像字节直接嵌入在image列中,实现了自包含,不依赖外部文件路径。syncred_600子集保留了来自源文件SynCred_600/meta.json的注释信息;fp_450子集则仅包含可从文件名可靠推导出的元数据。在数据导出过程中,源元数据中存在的subtyoe拼写错误已被规范化为subtype。该数据集适用于需要图像及其关联元数据的任务,例如图像分类、元数据分析和计算机视觉研究。数据规模在1,000到10,000个样本之间,采用Apache 2.0许可证。
Syncred-Bench is an image dataset comprising two subset configurations: syncred_600 (containing 600 sample rows) and fp_450 (containing 450 sample rows). The dataset is stored in Parquet format, with image bytes directly embedded in the `image` column, making it self-contained without relying on external file paths. The syncred_600 subset retains the annotation information from the source file SynCred_600/meta.json; the fp_450 subset only includes metadata that can be reliably inferred from filenames. During the data export process, the spelling error of "subtyoe" present in the source metadata has been standardized to "subtype". This dataset is suitable for tasks requiring images and their associated metadata, such as image classification, metadata analysis, and computer vision research. The dataset has a scale between 1,000 and 10,000 samples and is released under the Apache 2.0 license.
数据集概述
名称:Syncred-Bench
许可协议:Apache-2.0
数据集规模:1K < n < 10K 行
配置与子集
- syncred_600:包含 600 行数据,保留了来自
SynCred_600/meta.json的原始标注信息。 - fp_450:包含 450 行数据,仅包含可从文件名可靠推导出的元数据。
说明
- 导出的 Parquet 文件是自包含的:图像字节已嵌入到
image列中,而非指向本地文件系统路径。 - 源元数据中存在的拼写错误
subtyoe在导出过程中已被规范化为subtype。
加载示例
python from datasets import load_dataset
syncred = load_dataset("thu-coai/Syncred-Bench", "syncred_600", split="train") fp = load_dataset("thu-coai/Syncred-Bench", "fp_450", split="train")




