Skip to content

Dataset · bghira

Images with concepts and captions

bghira/photo-concept-bucket

A large dataset of captioned images under a free license for training generative models.

What's inside

The dataset is designed as a free-to-use, high-quality image dataset without attribution requirements and with an open license. All images are automatically captioned: captioning was performed on a GPU cluster including 38x3090, 6x4090, 8xA5000, and 2xA200 (Note: text says A100). Transformers and Accelerate were used in fp8 precision; the data includes images, tabular metadata, and text.

What you can do with it

  • Text-to-image model fine-tuning
  • LoRA training for image generation
  • Preparing data for image captioning
Size
100K–1M rows
Downloads this month
1.8K
Likes
74

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("bghira/photo-concept-bucket")