examples/NOTICE.md

Example data sources

The project code license does not replace these datasets' original terms. The bundled samples contain RGB images and, for the calibrated examples, camera matrices. They contain no depth maps, meshes, labels, or training metadata.

Hypersim

The hypersim_provided and hypersim_da3_pose samples come from the public Hypersim dataset, scenes ai_004_009 and ai_006_001, respectively, camera trajectory cam_00. Hypersim data is licensed under Creative Commons Attribution-ShareAlike 3.0. The images are tone-mapped RGB derivatives, re-encoded without embedded metadata. Their resolution and pixel values are preserved from the project-page input files. The supplied camera arrays use OpenCV camera-to-world coordinates and pixel intrinsics matching the image files. Frame identities are in data/manifest.json.

Please cite Mike Roberts et al., Hypersim: A Photorealistic Synthetic Dataset for Holistic Indoor Scene Understanding, ICCV 2021.

DTU

The dtu_provided sample comes from scan 24 of the DTU MVS dataset. These are rectified, resized/cropped RGB images in the common MVSNet data format (640 × 512), as used in the project-page case. Supplied pixel intrinsics match these processed files. PNG metadata has been removed without changing pixels. DTU data remains subject to the original dataset terms, independently of the InfiniSplat project license. The dataset page describes the data as freely available and requests citation of its accompanying paper.

Please cite Rasmus Jensen, Anders Dahl, George Vogiatzis, Engin Tola, and Henrik Aanæs, Large Scale Multi-view Stereopsis Evaluation, CVPR 2014.