max planck institut informatik

MPI Light Field Archive

This archive consists of nine synthetic and captured five real-world scenes. The database consists of scenes spanning large variety of different conditions such as: specular and transparent surfaces, thin and irregular structures, multiple light sources, multiple diffuse objects, day and night lights, indoor and outdoor, High contrast color and texture, highly reflecting surfaces and shadows, glossy surfaces and simple objects. For capturing realworld scenes, we used a one-meter long motorized linear stage with Canon EOS 5D Mark II camera and 50mm and 28mm lenses. After capturing all views we performed lens distortion correction using PTLens, estimated the camera poses using Voodoo camera tracker, and rectified all the images. For rendered images, we used cameras with off-axis asymmetric frustums. We set the zero-disparity plane such that one-third of a scene is located in front of the screen. For real-world scenes, the same effect was achieved by applying horizontal shift to the individual views. All the light fields are of identical spatial and angular resolution (960 X 720 X 101). The angular resolution was chosen high enough to avoid visible angular aliasing. This was achieved by assuring that the maximum on-screen disparities between consecutive view are around 1 pixel. To guarantee comfortable viewing, we limit the total disparity range to 0.2 visual degree.

We considered typical light field distortions that are specific to transmission, reconstruction, and display. For each distortion, we generated multiple light fields by varying the distortion severity level. Please refer to the paper and supplimentary material for more details.

Citation

Vamsi Kiran Adhikarla, Marek Vinkler, Denis Sumin, Rafał Mantuik, Karol Myszkowski, Hans-Peter Seidel, Piotr Didyk
Towards a quality metric for dense light fields
Proceedings of the IEEE Conf. on Computer Vision and Pattern Recognition (CVPR)


@inproceedings{Vamsi2017,
 author = {Vamsi Kiran Adhikarla and Marek Vinkler and Denis Sumin and Rafał Mantiuk and Karol Myszkowski and Hans-Peter Seidel and Piotr Didyk},
 booktitle = {Proceedings of the {IEEE} Conf. on Computer Vision and Pattern Recognition ({CVPR})},
 title = {Towards a Quality Metric for Dense Light Fields},
 year = {2017}
}

Download all referene light fields

Download all distorted light fields