Laval PanDORA HDR Dataset

Université Laval

Real indoor scenes with HDR ground truth — captured with dual 360° cameras

About

We propose the Laval PanDORA HDR Dataset, which accompanies the paper PanDORA: Casual HDR Radiance Acquisition for Indoor Scenes. The dataset comprises 14 real-world indoor environments acquired using a portable rig equipped with two synchronized 360° cameras simultaneously recording at two exposure settings. For each scene, the dataset provides dual-exposure panoramic video frames, validity masks excluding the camera operator and the secondary camera, structure-from-motion reconstructions, and reference, unsaturated HDR panoramas obtained using an exposure-bracketing acquisition pipeline.

The PanDORA capture apparatus — two synchronized 360° cameras mounted on a portable monopod
The PanDORA capture apparatus: two synchronized 360° cameras recording simultaneously at two exposure levels.
14Scenes
2Camera streams
2Exposure levels
195GT HDR panoramas (EXR)
3,509Input frames per camera

Dataset Structure

Each scene folder follows the layout below. Structure-from-motion reconstruction was performed using OpenSfM.

<scene>/
├── data/
│   ├── left_e1/          # Left camera, regular exposure (panoramic frames)
│   ├── left_sfm/         # Left camera, frames for self-calibration
│   ├── right_sfm/        # Right camera, frames for self-calibration
│   ├── right_e2/         # Right camera, fast exposure (panoramic frames)
│   ├── masks/            # Binary masks (per camera stream)
│   │   ├── left_e1/
│   │   ├── left_sfm/
│   │   ├── right_e2/
│   │   └── right_sfm/
│   └── capture_settings.json
├── GT/
│   ├── GT_exr/           # HDR ground-truth panoramas (.exr + .png previews)
│   └── GT_exr_renders/   # Rendered HDR views (.exr + .png previews)
└── sfm/
    └── reconstruction.json
PathDescriptionFormat
data/left_e1/Left 360° camera, regular exposure — input framesPNG
data/left_sfm/Left camera frames used for self-calibrationPNG
data/right_sfm/Right camera frames used for self-calibrationPNG
data/right_e2/Right 360° camera, fast (dark) exposure — input framesPNG
data/masks/Binary masks aligned to each camera streamPNG
data/capture_settings.jsonCamera exposure and capture parametersJSON
GT/GT_exr/HDR ground-truth panoramas (one per GT viewpoint)EXR + PNG
GT/GT_exr_renders/Rendered perspective views from GT panoramasEXR + PNG
sfm/reconstruction.jsonOpenSfM reconstruction. See OpenSfM format docs.JSON

Data Preview — Auditorium

Sample frames from each data stream for the Auditorium scene. Left and right columns share the same timestamps.

Left camera
(regular exposure)
Left e1 t=0010 Left e1 t=0780 Left e1 t=1550
Left binary mask
Left mask t=0010 Left mask t=0780 Left mask t=1550
Right camera
(fast exposure)
Right e2 t=0010 Right e2 t=0780 Right e2 t=1550
HDR ground-truth renders
GT render 1 GT render 8 GT render 15

Scenes

To download a scene, click on its image.

Lab office Lab office GT
Lab office
930 input frames18 GT HDR views
Meeting room Meeting room GT
Meeting room
510 input frames15 GT HDR views
Auditorium-dark Auditorium-dark GT
Auditorium-dark
326 input frames15 GT HDR views
Auditorium Auditorium GT
Auditorium
384 input frames15 GT HDR views
Blue bedroom Blue bedroom GT
Blue bedroom
308 input frames15 GT HDR views
Classroom-windows Classroom-windows GT
Classroom-windows
1,550 input frames15 GT HDR views
Classroom-no windows Classroom-no windows GT
Classroom-no windows
708 input frames15 GT HDR views
Coffee room Coffee room GT
Coffee room
302 input frames15 GT HDR views
Basement Basement GT
Basement
516 input frames15 GT HDR views
Office Office GT
Office
406 input frames15 GT HDR views
Clubhouse Clubhouse GT
Clubhouse
880 input frames15 GT HDR views
Living room Living room GT
Living room
822 input frames15 GT HDR views
Lobby Lobby GT
Lobby
552 input frames15 GT HDR views
Small office Small office GT
Small office
402 input frames12 GT HDR views
🕐 More scenes coming soon. Additional indoor environments are currently being processed and will be added to this dataset in future releases.

Code

The training and evaluation code for PanDORA is currently being prepared for public release. It will include the full NeRF-based pipeline for HDR radiance acquisition from dual-exposure panoramic video, along with preprocessing scripts and evaluation utilities.

📄 Code release coming soon. The source code will be made publicly available shortly. Stay tuned via the project page.

License

The Laval PanDORA HDR Dataset is released under a custom research license. By downloading or using this dataset, you agree to the following terms:

You are free to:
Use, process, and build upon this dataset for any purpose, including academic research, education, and commercial research, provided that you cite the original paper (see Citation below).

You may not:
Redistribute, re-host, mirror, or otherwise make this dataset publicly available from a source other than the official project page. Users must download the data directly from this page.

Attribution:
Any publication or work that uses this dataset must cite:
PanDORA: Casual HDR Radiance Acquisition for Indoor Scenes, Karimi Dastjerdi et al., arXiv:2407.06150.

Citation

If you use this dataset, please cite:

@misc{dastjerdi2025pandoracasualhdrradiance,
  title         = {PanDORA: Casual HDR Radiance Acquisition for Indoor Scenes},
  author        = {Mohammad Reza Karimi Dastjerdi and Dominique Tanguay-Gaudreau and Frédéric Fortier-Chouinard and Yannick Hold-Geoffroy and Claude Demers and Nima Kalantari and Jean-François Lalonde},
  year          = {2025},
  eprint        = {2407.06150},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2407.06150}
}

Contact

For questions about the dataset, feel free to reach out to: