Omnidirectional HDR consumer camera dataset

Dataset

Omnidirectional representations enable multimedia content consumers to experience immersion and engagement. However, Standard Dynamic Range (SDR) displays used to visualize these contents limits visual signal naturalness and realism.
We propose here to improve the realism of 360˚ contents by exploiting High Dynamic Range (HDR) representations.

The 360˚ HDR dataset presented here is composed by 43 contents. Each content representation includes a set of 5 multi-exposure pictures, a HDR reconstruction as well as tone-mapped and exposures fusion versions of the scene.

The multi-exposure pictures were captured using the consumer camera Ricoh Theta S, with EVs ranging from -2 to 2 by step of 1. Each picture acquired by the Ricoh Theta S is 8 bit RGB 5476×2688 (14MB) and was compressed with a JPEG quality parameter of 95.

The HDR reconstruction is generated following the work of Debevec and Malik.
The four off-the-shelf Tone Mapping Operators (TMOs) used to mapping HDR representation to SDR are the linear TMO, Photographic TMO, Display adaptive TMO and Detail preserving TMO. The Exposure fusion algorithm is a state of the art algorithm. All the used operators are described in the Table 1.

The captured contents are classified in three content types, namely Indoor  (9 contents), Outdoor  (24 contents) and Night  (10 contents).

For more information, please read the work of Perrin and al. mentioned hereinafter.

Ratings, tracking and toggling information gathered during Perrin and al. work are also included.

Permission is hereby granted, without written agreement and without license or royalty fees, to use, copy, modify, and distribute the data provided and its documentation for research purpose only. The data provided may not be commercially distributed. In no event shall the Ecole Polytechnique Fédérale de Lausanne (EPFL) be liable to any party for direct, indirect, special, incidental, or consequential damages arising out of the use of the data and its documentation. The Ecole Polytechnique Fédérale de Lausanne (EPFL) specifically disclaims any warranties. The data provided hereunder is on an “as is” basis and the Ecole Polytechnique Fédérale de Lausanne (EPFL) has no obligation to provide maintenance, support, updates, enhancements, or modifications.

If you use these database in your research we kindly ask you to reference the associated paper:

Perrin, A-F., Bist, C., Cozot,R., and Ebrahimi, T. Measuring quality of omnidirectional high dynamic range content. SPIE, San Diego, California, United States, August 6-10, 2017.

Contact

Should you have any questions regarding this research please contact Anne-Flore Perrin ([email protected])

Download

You can download all the dataset ( set of 5 multi-exposures pictures, the HDR reconstruction as well as tonne-mapped and exposure fusion versions of a scene) from the following ftp (please use dedicated FTP clients, such as fileZilla or FireFTP).

FTP address: tremplin.epfl.ch
Password: ohsh9jah4T

The dataset can be found in the folder HDR_360. The total size of the dataset is about 4.3 GB.