LiFF Light Field Features Toolbox

This is an implementation of the light field feature detector and descriptor from the paper:

D. G. Dansereau, B. Girod, and G. Wetzstein, “LiFF: Light field features in scale and depth,” in Computer Vision and Pattern Recognition (CVPR), 2019. Available here, supplementary material here.

The functionality closely mirrors the SIFT feature detector and descriptor, but exploits information in the light field to deliver more robust and informative features.

This works by searching for features with well-defined scale, as in SIFT, but also with well-defined depth, which manifests as slope in the light field. This rejects more spurious features, detects partially occluded features, and builds descriptors with greater immunity to partial occlusions and higher-order light transport effects. It also delivers a per-feature depth (slope) estimate, and runs in 2-3 seconds on a Lytro Illum light field, about 18x faster than repeating SIFT across the light field.

Download

Latest release at the LiFF GitHub page
Sample Scenes: SampleScenes.tar
Multiview Datasets at Stanford and supplementary material here
More examples in ESLF format at Stanford