Many papers report results for home videos (see, e.g., M. Del Fabro and L. Böszörmenyi, Multimedia Systems 19:427-454 2013), but only a few datasets are publicly available. In the following we have collected the datasets that have been found and commented their quality. The preferred order of use is:

USAA

1. CCV (sufficiently large, Youtube, recent and public)
2. SumMe? (sufficiently large, Youtube, recent and public)
3. CASA2 (sufficiently large, selflog with good audio, recent, not public)

• Download page

USAA and Kodak video datasets are outdated and actually CCV replaces them.

SumMe?

• Database Web site

CASA2

Collected by the SGN Audio Group

• Audio event detection demo using CASA2

USAA

• Download page

-- JoniKaemaeraeinen - 21 May 2014

Urban City Behavior

These datasets can be used to detect and analyze human behavior in urban city areas (people walking, crowd counting etc.)

LOST

Database Web site

-- JoniKaemaeraeinen - 16 Jul 2015