BiggieW/DataComp-12M
收藏Hugging Face2026-03-25 更新2026-03-29 收录
下载链接：
https://hf-mirror.com/datasets/BiggieW/DataComp-12M
下载链接
链接失效反馈官方服务：
资源简介：
---
dataset_info:
features:
- name: uid
dtype: string
- name: url
dtype: string
- name: ori_text
dtype: string
- name: syn_text
sequence: string
license: cc-by-4.0
task_categories:
- zero-shot-image-classification
- text-to-image
- image-to-text
tags:
- datacomp
- clip
---
# DataComp-12M: Merged Metadata
This dataset provides a unified metadata view of [DataComp-12M](https://huggingface.co/datasets/mlfoundations/DataComp-12M) and [DataCompDR-12M](https://huggingface.co/datasets/apple/DataCompDR-12M), containing
image URLs, original captions, and synthetic captions in a compact parquet format (~3.6GB total).
## Dataset Description
Each row corresponds to one image-text sample from the DataComp-12M pool, enriched with synthetic captions from DataCompDR-12M. The two datasets share the same UIDs.
| Column | Type | Description |
|--------|------|-------------|
| `uid` | `string` | Unique sample identifier (shared across DataComp-12M and DataCompDR-12M) |
| `url` | `string` | Original image URL |
| `ori_text` | `string` | Original alt-text caption (from DataComp-12M) |
| `syn_text` | `list[string]` | Synthetic captions (from DataCompDR-12M) |
提供机构：
BiggieW
