Hugging Face's logo

Datasets:
Databoost
/
sophistasia

Modalities:
Image
Text
Formats:
parquet
Size:
< 1K
Libraries:
Datasets
pandas
Dataset card Viewer Files Community
1
sophistasia / README.md
Databoost's picture
Databoost
Update README.md
7b2690a
preview code
|
raw
history blame contribute delete
2.47 kB

Documentation of Sophistasia

Dataset Summary

This repository contains an annotated dataset of 995 images, categorized into 268 Landscape, 341 Culture, and 386 Street. Each image is accompanied by a description, location, and category. The dataset supports tasks such as image captioning, image-text retrieval, and text-to-image generation, making it a valuable resource.

Thematic Chosen

This dataset includes three thematic categories:

  • Landscape: Focuses on natural and architectural elements.
  • Culture: Highlights traditions and social practices.
  • Street: Covers urban environments and public spaces.

These themes provide a comprehensive and diverse dataset.

Supported Tasks

  • Image Captioning: The dataset can be used to train a model that automatically generates a description from an image.
  • Image-Text Retrieval: It can also be used for multimodal search models, linking images and text to retrieve one based on the other.
  • Text-to-Image Generation: The dataset enables training a model that creates images from textual descriptions.

File Format

The dataset is provided in the Parquet format, ensuring efficient storage and access.

Organisation of the Project

The project is organized under the main directory Sophistasia. Inside, there is a data folder containing a train subfolder with the data.parquet file. The images folder is divided into two sections:

  • brutes (with subfolders culture, landscape, street for the original images)
  • resized (containing the same subfolders for the resized images).

A readme.md file is also included with project information.

Field Description

The descriptions of the fields are as follows:

  • image: A string that represents the filename of the image listed in the "image" folder
  • caption: A string containing a concise description of the scene depicted in the image.
  • Location: A string specifying the original location where the image was taken or what it represents.
  • category: A string indicating the category to which the image belongs.

Image Sources

The images in this dataset were either:

  • Captured by our team members.
  • Sourced from copyright-free platforms like Pexels, ensuring they are royalty-free and available for use.

Contact

For any inquiries, please reach out to: