Skip to content

Dataset · pixparse

Web images with Conceptual Captions (CC3M) descriptions

pixparse/cc3m-wds

Classic corpus of ~3.3 million image-text pairs collected from HTML Alt-text tags.

What's inside

A compilation of the popular Conceptual Captions (CC3M) dataset containing 1 to 10 million examples (about 3.3 million pairs originally). The data was collected from the internet based on web image Alt-text attributes, featuring a wide variety of visual styles compared to manually annotated sets.

What you can do with it

  • Training and fine-tuning image-to-text caption generation models
  • Pre-training multimodal encoders and text-to-image retrieval systems
  • Training LoRA adapters for understanding diverse web content
Tasks
Image → text
Size
1M–10M rows
Downloads this month
19.5K
Likes
61

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("pixparse/cc3m-wds")