EgoSampling: Fast-Forward and Stereo for Egocentric Videos

Input Videos for Fast-Forward Experiments

The table below lists the sequences we used in our experiments. For each input sequence, we provide the result of a naive fast-forward (uniform sampling) and our method (both 1st and 2nd order smoothness). All results are given both before stabilization ('Raw') and after stabilization using YouTube's stabilizer ('Stabilized').

Input Sequence	Start/End Frames	x10 Naive Fast-Forward	Our Output 1st Order	Our Output 2nd Order
Bike1 [1]	350→11136	[Raw] [Stabilized]	[Raw] [Stabilized]	[Raw] [Stabilized]
Bike2 [1] (Same file as Bike1)	16150→23199	[Raw] [Stabilized]	[Raw] [Stabilized]	[Raw] [Stabilized]
Bike3 [1]	5800→29500	[Raw] [Stabilized]	[Raw] [Stabilized]	[Raw] [Stabilized]
Walking1 [1]	2800→20049	[Raw] [Stabilized]	[Raw] [Stabilized]	[Raw] [Stabilized]
Walking2 [2]	700→7600	[Raw] [Stabilized]	[Raw] [Stabilized]	[Raw] [Stabilized]
Running1	2100→15000	[Raw] [Stabilized]	[Raw] [Stabilized]	[Raw] [Stabilized]
Driving2	1800→12000	[Raw] [Stabilized]	[Raw] [Stabilized]	[Raw] [Stabilized]

References:

[1] Johannes Kopf,Michael Cohen, Richard Szeliski, First-person Hyperlapse Videos - Supplemental Material. http://research.microsoft.com/enus/um/redmond/projects/hyperlapse/supplementary/index.html.

[2] Alireza Fathi, Jessica K. Hodgins, James M. Rehg, Social Interactions: A First-Person Perspective, IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), 2012.