ULB Bunny dataset

ULB Bunny Dataset

Description

The ULB Bunny Dataset is an 80-view 360° multiview Blender rendering of a metallic Stanford Bunny placed in a Lambertian scene.  This dataset is intended for research on non-Lambertian surface reconstruction, Gaussian Splatting, and material-aware rendering.  All views are rendered under controlled lighting to ensure consistency and reproducibility.

---

Contents

The dataset includes:

- 80 PNG images    High-resolution renders from uniformly sampled viewpoints covering a full 360°.

- COLMAP reconstruction files   Provided in the Gaussian Splatting–compatible format.

Directory structure:

-- images/     *.png  (80 images)

-- sparse/     -- 0/          cameras.bin          images.bin          points3D.bin

Format

Images: PNG, undistorted, one image per camera viewpoint

Camera calibration: Stored in cameras.bin

Camera poses: Stored in images.bin

Sparse point cloud: Stored in points3D.bin

The structure follows the standard COLMAP sparse model expected by most Gaussian Splatting implementations.

Intended Use

This dataset is suitable for:

Evaluating non-Lambertian surface reconstruction

Testing 3D Gaussian Splatting and material-aware Gaussian variants

Benchmarking differentiable rendering algorithms

Studying view-dependent effects under controlled conditions

Citation

If you use this dataset, please cite:

Dubar, E., Dury, S., Bonatto, D., & Lafruit, G. (2025). ULB Bunny Dataset. Zenodo. https://doi.org/10.5281/zenodo.17750680

@dataset{ulb_bunny_dataset_2025,  author       = {Eva Dubar and Sarah Dury and Daniele Bonatto and Gauthier Lafruit},  title        = {ULB Bunny Dataset},  year         = {2025},  publisher    = {Zenodo},  doi          = {10.5281/zenodo.17750680},  url          = {https://doi.org/10.5281/zenodo.17750680}}