The effective propagation of pixel labels through the spatial and temporal domains is vital to many computer vision and multimedia problems, yet little attention has been paid to the temporal/video domain propagation in the past. We have begun to explore this problem in the context of general real-world "videos in the wild" and surveillance videos. Our current efforts primarily focus on mixing motion information with appearance information Previous video label propagation algorithms largely avoided the use of dense optical flow estimation due to their computational costs and inaccuracies, and relied heavily on complex (and slower) appearance models.

We used a subset of the videos from xiph.org as the basis of our benchmark dataset for label propagation. Existing datasets either restricted the study to two classes or were taken in restricted settings, such as from the dash of a moving vehicle. Our new data set has general motion and presents stratified levels of complexity. We continue to add to the labels and will release additional videos in the future. For more information, see Albert Chen's page.

Download the full data set.

Code from our WNYIPW paper is here with the config file. Or you can get the dataset above and the code is in the package.
If you use the dataset or the code, the associated cite is below.

This work is partially support by NSF CAREER IIS-0845282 [project page].