trevorcanham/hdrvs_adaCGsubset

High Dynamic Range Video Dataset (HDRVS) Adaptive Color Grading Subset

This is the dataset subset of HDRVS used in Adaptive Color Grading. The data are 1564 still images mined from a large dataset of 784 RAW videos (2 stills for video) and processed for viewing in DisplayP3 gamma 2.4. The data are annotated in the associated .pkl file with 4 subjective thresholds for darkest, dark, light and lightest image regions using the color grading tool from the above repository. Using this repository, the file can be opened as a project file, or directly applied to images using the associated submodules. The annotation dictionary trainDemo8b.pkl has the following form:

{
  'root': 'C:/path/to/images',          # where the images live
  'IMG_0001.tif': {                     # key: path relative to root
      'set':    {'darkest': [a, b], 'dark': [a, b], 'light': [a, b], 'lightest': [a, b]},
      'reg':    {'darkest': [pivot, falloff], ...},
      'hist':   np.ndarray,             # luminance histogram, the adapt input feature
      'thumb':  np.ndarray,             # JPEG byte stream, decoded on demand
      'labels': ['subsetA', ...],       # free-form tags, filterable
      'time':   datetime,
  },
  ...
}

If used please cite the following:

@misc{hdrvs2026,
  title  = {The HDR Videographic Survey},
  year   = {2026},
  howpublished = {\url{https://hdrvs.github.io/}}

@inproceedings{acg,
  title={Adaptive Color Grading},
  author={Canham, Trevor and Punnappurath, Abhijith and Brown, Michael},
  booktitle={Color and Imaging Conference},
  year={2026}
}