... consistency. Additional meta information allows for a precise selection of testing data, as well as training andevaluation of recognition systems. The meta tags for eachclip include the camera view-point, ... internet-scale static image databases [17, 23, 27, 4, 5], current ac-tion recognition datasets lag far behind. The most popularbenchmark datasets, such as KTH [20], Weizmann [3] or theIXMAS dataset ... recogni-tion rate. In this context, we describe an effort to advancethe field with the design of a large video database contain-ing 51 distinct action categories, dubbed the Human Mo-tion DataBase (HMDB51),...