Hugging Face's logo

Datasets:
jingchao-peng
/
HDRTDataset

Tasks:
Image-to-Image
Languages:
English
Size:
10K<n<100K
ArXiv:
Tags:
HDR
RGB-T
Infrared
License:
Dataset card Files Community

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

HDRT: A Large-Scale Dataset for Infrared-Guided HDR Imaging

Graphical Abstract

Overview

The HDRT dataset is a large-scale dataset designed for infrared-guided high dynamic range (HDR) imaging. It includes aligned infrared (IR), standard dynamic range (SDR), and HDR images to facilitate research in multi-modal fusion, HDR imaging, and any other related research.

Files

  • hdr_reduced_size.zip contains HDR data with a resolution of 640×480. Files with the suffix .hdr are true high dynamic range images saved in floating point numbers. Files with the suffix .JPG are 8-bit images generated by Canon cameras that combine images with different exposures. They can be regarded as high dynamic range images after tone mapping.

  • hdr_4k.zip (including hdr_4k.z01 ~ hdr_4k.zip) contains HDR images with a resolution of 5120×3840. The file name rule is same as hdr_reduced_size.zip.

  • registered_RGB_reduced_size.zip contains SDR images with different exposures with a resolution of 640×480. For every 4 consecutive images (e.g. RGB00261.JPG, RGB00262.JPG, RGB00263.JPG, RGB00264.JPG), The first image (RGB00261.JPG) is an SDR image with a well-exposure value. The second image (RGB00262.JPG) is an SDR image with an under-exposure value. The third image (RGB00263.JPG) is an SDR image with an over-exposure value. The fourth image (RGB00264.JPG) is created by Canon by fusing the above three images.

  • registered_RGB_4K.zip (including registered_RGB_4K.z01 ~ registered_RGB_4K.zip) contains SDR images with different exposures with a resolution of 5120×3840. The file name rule is same as registered_RGB_reduced_size.zip.

  • infrared.zip contains IR images (640×480) in TIFF format starting with T+id (e.g. T00261.tiff). Each id is the same as the corresponding first id from the RGB folder. For example, there are SDR images with file names RGB00261.JPG, RGB00262.JPG, RGB00263.JPG, and RGB00264.JPG; HDR images with file names RGB00261.hdr and RGB00261.JPG; the corresponding infrared image is T00261.tiff.

  • infrared_1280_960.zip contains IR images (1280×960) in JPEG format with the same file name rule as infrared.zip.

Dataset Composition

Category Count Resolution
HDR Images 10,000 640×480 and 5120×3840
SDR Images 30,000 640×480 and 5120×3840
IR Images 10,000 640×480 and 1280×960

Sample Scenes

  • Urban and rural landscapes

  • Natural scenery, trees, and parks

  • Vehicles, pedestrians, and buildings

  • Wellexposed, overexposed, and underexposed conditions

Data Collection

  • RGB Camera: Canon EOS 5D Mark III

  • IR Camera: Optris PI 640i (640×480)

  • Capture Method: Synchronized RGB and IR image collection

  • Post-processing: Manual registration for precise alignment

Citation

If you use the HDRT dataset, please cite our work:

@misc{HDRT2024,
  title={HDRT: Infrared Capture for HDR Imaging},  
  author={Jingchao Peng and Thomas Bashford-Rogers and Francesco Banterle and Haitao Zhao and Kurt Debattista},
  year={2024},
  eprint={2406.05475},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2406.05475}, 
}
Downloads last month
35