Skip to main content

Table 4 Comparison results of the proposed with other state-of-art action recognition methods on HMDB51

From: Consistent constraint-based video-level learning for action recognition

Training method

Method

Accuracy (%)

Clip level

Geometry [32]

23.3

 

MASN [23]

32.6

 

ST-puzzle (Kinetics) [35]

28.3

 

MASN (Kinetics) [23]

33.4

Video level

Ours

35.38