Dataset Card for pxhere Images

Dataset Summary

This dataset contains a large collection of high-quality photographs sourced from pxhere.com, a free stock photo website. The dataset includes approximately 760,600 images in full resolution covering a wide range of subjects including nature, people, urban environments, objects, animals, and landscapes. All images are provided under the Creative Commons Zero (CC0) license, making them freely available for personal and commercial use without attribution requirements. The dataset is organized in WebDataset format, making it suitable for large-scale machine learning applications such as image classification, image-to-text generation, and computer vision research.

Languages

The dataset is primarily monolingual:

• English (en): Image titles, tags are predominantly in English.

Dataset Structure

Data Files

The dataset consists of:

• Images and metadata organized in WebDataset format (.tar files)
• 761 archive files, each containing about 1,000 images
• File naming convention: pxhere-000000.tar through pxhere-000760.tar
• Each archive contains images in their original, full resolution along with corresponding JSON metadata

Data Fields

Each image in the dataset is accompanied by a JSON metadata file containing the following fields:

• image_id: Unique identifier for the image on PxHere.com, also used as the filename.
• download_url: The original URL to download the image from PxHere.com.
• tags: Array of strings representing categories and descriptive terms associated with the image.
• uploaded: The upload date of the image in MM/DD/YYYY format.
• exif_info: Object containing EXIF metadata from the original photo (when available):
  • camera_model: The camera used to take the photograph.
  • resolution: The image dimensions in pixels (width × height).
  • aperture: The f-stop setting used.
  • focal_length: The focal length of the lens used.
  • shutter_speed: The exposure time.
  • iso: The ISO sensitivity setting.
  • lens_details_raw: Combined lens and exposure details.

Data Splits

The dataset contains a single split:

Split	Number of Examples
train	≈760,600

Total dataset size: ≈760,600 entries

Data Format

• Images are stored in their original format (typically JPEG) at full resolution
• Each image is named according to its unique image_id
• Metadata is stored in corresponding JSON files with the same base name as the image
• All files are packaged in WebDataset format (tar archives), with approximately 1,000 images per archive

License Information

Licensing Structure

All images in this dataset are distributed under the Creative Commons Zero (CC0) license, which effectively places them in the public domain. This means:

• No attribution is required for use
• Images can be used for both personal and commercial purposes
• Images can be modified and redistributed without restrictions
• No royalties or payments are required for any use

CC0 License Details

The Creative Commons Zero (CC0) license waives all copyright and related rights to the work worldwide. Users can copy, modify, distribute, and perform the work, even for commercial purposes, without asking permission.

For detailed information about the CC0 license, please refer to: https://creativecommons.org/publicdomain/zero/1.0/