BounceBRDF Demo Dataset

Demo dataset for the paper "High-Gloss SVBRDF Capture Using Bounce Light" (Eurographics 2026 / Computer Graphics Forum, Vol. 45, No. 2).

Authors: Tomáš Iser, Andrei-Timotei Ardelean, Tim Weyrich — Friedrich-Alexander-Universität Erlangen-Nürnberg & Charles University

📄 Project page & paper  ·  💻 GitHub Source Code (BounceBRDF)

Contents

Four captured scenes (arch1, re2, vr1, vr3), each containing:

<scene>/
├── IMG_XXXX.exr            # RAW input photos (with mirror sphere) in EXR format
├── IMG_XXXX.envmap.exr     # Extracted environment map for each photo (using extract_envmap.py)
└── results/                # Results obtained from a subset of 9 input photos & 1000 iterations
    ├── basecolor.png       # Base color texture
    ├── roughness.png       # Roughness texture
    ├── metallic.png        # Metallic texture
    └── normal.png          # Normal map

The input photos and environment maps are in an EXR (.exr) image format, which is an open standard for storing camera captures with high dynamic range. The EXR images were exported from the camera RAW files using darktable. The environment maps were extracted from the mirror sphere using our extract_envmap.py script. The four results/ textures are the SVBRDF output of our fit.py script. See the source code on GitHub.

Usage

To reproduce the results or run the pipeline on your own data, see our BounceBRDF repository. The extracted environment maps are already provided, so you can skip the extract_envmap step and run fit directly:

python -m bouncebrdf.fit '<scene>/*.exr'

Citation

@article{iser2026highgloss,
  author    = {Iser, Tom\'{a}\v{s} and Ardelean, Andrei-Timotei and Weyrich, Tim},
  title     = {High-Gloss {SVBRDF} Capture Using Bounce Light},
  journal   = {Computer Graphics Forum (Proc. Eurographics)},
  volume    = {45},
  number    = {2},
  year      = {2026},
  month     = may,
  publisher = {Eurographics Association},
  authorurl = {https://reality.tf.fau.de/pub/iser2026highgloss.html},
}