---
license: cc-by-4.0
task_categories:
  - image-to-3d
tags:
  - physical-ai
  - nurec
  - 3d-reconstruction
  - novel-view-synthesis
  - radiance-fields
  - photometric-calibration
pretty_name: PPISP Dataset
size_categories:
  - 1K<n<10K
---

# PPISP Dataset

## Dataset Description:
The PPISP dataset accompanies the work "PPISP: Physically-Plausible Compensation and Control of Photometric Variations in Radiance Field Reconstruction". It contains object-centric scene captures of four outdoor scenes, each captured with three different cameras, for multi-view 3D reconstruction and novel view synthesis. The photos were captured with exposure bracketing of +/-2 EV and re-processed with automatic exposure and color corrections to create a challenging benchmark for methods that compensate for photometric inconsistencies.

Each scene has two variants: a standard version processed from the full exposure brackets, and an auto version where selected exposures were re-processed with automatic exposure compensation and white balancing, for a total of eight sequences. We also provide pre-computed COLMAP sparse reconstructions (camera poses and point clouds) for each sequence, enabling direct usage in common radiance field reconstruction methods.

This dataset is ready for commercial/non-commercial use.

## Dataset Owner(s):
NVIDIA Corporation

## Dataset Creation Date:
22 January 2026

## License/Terms of Use: 
This dataset is licensed under the [Creative Commons Attribution 4.0 International License (CC-BY-4.0)](https://creativecommons.org/licenses/by/4.0/).

## Intended Usage:
Benchmarking and development of radiance field reconstruction methods such as 3DGRUT and GSplat.

## Dataset Characterization
**Capture**

Four scenes were captured in raw format using two mirrorless camera systems (Nikon Z7, OM System OM-1 Mark II) and a smartphone (Apple iPhone 13 Pro). The captures were done in an object-centric fashion, that is, with a certain object (statue, artwork, etc.) as the main focus, with photos taken from multiple angles around it. Exposure bracketing of +/-2 EV was used. Aperture and focus were set manually and remained fixed, and image stabilization was disabled.

**Processing**

The raw photos were developed to JPEG using the manufacturer's recommended software (NX Studio for Nikon, OM Workspace for OM System, Adobe Lightroom Classic for iPhone), with manual white balance from a color calibration target. Furthermore, selected exposures were re-processed with automatic exposure compensation and white balancing, producing an extra four sequences for a total of eight sequences from four real-world scenes.

We used COLMAP to reconstruct camera poses and a sparse point cloud for each sequence.

## Dataset Format
Downsampled images in JPEG format, COLMAP camera poses, COLMAP sparse point cloud.

## Dataset Quantification
8 sequences, about 2600 photos in JPEG format total. The four standard sequences contain 500-700 images each; the four auto sequences contain 50-150 images each.

Measurement of Total Data Storage: About 7.5 GB in compressed form.

## Reference(s):
[Project Page](https://research.nvidia.com/labs/sil/projects/ppisp/)

## Ethical Considerations:
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications.  When downloaded or used in accordance with our terms of service, developers should work with their internal team to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse.   
Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns here.