jadgardner/reni-hdr

RENI HDR

RENI HDR is a curated dataset of outdoor, natural high-dynamic-range illumination environments used to train and evaluate RENI and RENI++. Each image is a full equirectangular environment map in linear RGB radiance.

The source panoramas were obtained under CC0 and manually curated. Images containing personally identifiable information, offensive content, or predominantly unnatural light sources were removed.

Contents

The archive extracts to the directory expected by ns_reni:

RENI_HDR/
  train/       # 1,673 EXRs at 128 x 64
  val/         # 10 high-resolution EXRs at 1024 x 512
  test/        # 21 held-out EXRs at 128 x 64
  masks/       # five fixed 512 x 256 completion masks

The test images are held out from decoder training and are used by fitting only their latent codes. The validation maps retain their high-resolution versions for qualitative evaluation; the loader resamples them when a fixed query resolution is requested.

This release deliberately excludes the working directory's later pseudo-sun labels, Blender files, object meshes, inverse-rendering renders and unrelated image assets. Those are not part of the curated illumination corpus.

Download

Download and verify the tagged release:

hf download jadgardner/reni-hdr \
  --repo-type dataset \
  --revision v1.0 \
  --local-dir reni-hdr

(cd reni-hdr && sha256sum -c SHA256SUMS)
tar --zstd -xf reni-hdr/archives/reni-hdr.tar.zst -C /path/to/data

The result is /path/to/data/RENI_HDR, which can be passed directly to RENI++:

ns-train reni --data /path/to/data/RENI_HDR

The archive can also be downloaded without the Hugging Face CLI:

curl -L \
  "https://huggingface.co/datasets/jadgardner/reni-hdr/resolve/v1.0/archives/reni-hdr.tar.zst?download=true" \
  -o reni-hdr.tar.zst

CONTENTS.json records the release-relative path, size and SHA256 of all 1,709 extracted files. DATASET_STATS.json records the validated split counts and resolutions. MANIFEST.json and SHA256SUMS cover the downloadable release files.

Provenance

The environments were curated from the CC0 sources listed in SOURCES.md. The collected files were sequentially renamed during the original curation, so a per-image source-name mapping is not available. This limitation is recorded explicitly rather than inferring provenance after the fact.

Licence

The dataset is released under the CC0 1.0 Universal Public Domain Dedication. See LICENSE.md.

Citation

Please cite the RENI and RENI++ publications:

@inproceedings{gardner2022reni,
  title     = {Rotation-Equivariant Conditional Spherical Neural Fields for
               Learning a Natural Illumination Prior},
  author    = {Gardner, James A. D. and Egger, Bernhard and
               Smith, William A. P.},
  booktitle = {Advances in Neural Information Processing Systems},
  volume    = {35},
  pages     = {26309--26323},
  year      = {2022}
}

@article{gardner2026renipp,
  title   = {{RENI++}: A Rotation-Equivariant, Scale-Invariant, Natural
             Illumination Prior},
  author  = {Gardner, James A. D. and Egger, Bernhard and
             Smith, William A. P.},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year    = {2026},
  doi     = {10.1109/TPAMI.2026.3691593}
}