Memory Efficient Stereoscopy from Light Fields - Supplementary Material

Figures S1–S4 show the complete set of input and output images of the four tasks we experimented using four different datasets at various resolutions. Since the Elephant dataset was available only at 1k resolution without depth information, we only conducted the user scribbles and the constant disparity tasks at 1k resolution for this dataset. Please also refer to our paper for more details of the experiments. Table S1 lists the root-mean-squared errors (RMSE) of our results against the ground truth for the constant disparity and linear scaling tasks, where the new view is computed with respect to a constant disparity and a linear scaling of the scene depth, respectively. We measured the error for two different resolutions, 1280×853 (1k) and 1920×1280 (2k). Refer to Figures S3 and S4 for the corresponding results.