Phase-Based Frame Interpolation for Video Supplementary Material


In Figure 1 we report error measures using the perceptually motivated structural similarity (SSIM) measure. This complements the sum of squared distances (SSD) error measures reported in Figure 8 (right) in the paper. 1 MDP-Flow2 SimpleFlow Brox et al. flowlib Pyramid LK Linear Blending Didyk et al. Our Method Figure 1: Error measurements (SSIM) for the different sequences shown in Figure 2. Note that a higher value is better with 1 being the maximum. In Figure 2 we show example input images from the sequences used to compute the error measures in Figure 1 as well as in Figure 8 (right) in the paper. In Figure 3 we compare our phase-based method to optical flow on the Middlebury dataset 1. In order to increase the visual quality of our results, we ignored the high pass residual in these examples, which, however, leads to larger numerical errors compared to the ground truth.racy optical flow estimation based on a theory for warping. (a) Barrier (b) Couple (c) Face (d) Hair (e) Handkerchief (f) Sand (g) Firemen (h) Light (i) Roto Figure 2: The sequences used for the error measurements in Figure 1 as well as in Figure 8 (right) in the paper.

3 Figures and Tables

Cite this paper

@inproceedings{Meyer2015PhaseBasedFI, title={Phase-Based Frame Interpolation for Video Supplementary Material}, author={Simone Meyer and Oliver Wang and Henning Zimmer and Max Grosse and Alexander Sorkine-Hornung}, year={2015} }