Deep Learning Based Dynamic Patterns Classification of Videos Using Spatio-Temporal Virtual Communication
S. P. Kasthuri Arachchi, Yi-Cheng Chen, Timothy K. Shih
National Central University, Taiwan (R. O. C.)
Misclassification Results of VC-LSTM Model
(a)
(b)
(c)
Sample misclassification results. Even figure (a) is a frame of a video in Crosette class, it has classified as WaterFlower class. Figure (b) is classified as in Palm class even it is in WaterFlower class. Besides, figure (c) recognized as Dasi class even it originally contains in Drop class; (a) sample frame of Crosette class; (b) sample frame of WaterFlower class; (c) sample frame of Drop class
Misclassification Results of Single-stream Model
(a)
(b)
(c)
(d)
(e)
(f)
(g)
Even Figure (a) and (b) are two frames of a particular video clip in the Desi class, it’s classified as Chrysanhemum class. By the same token, Figure (c) and (d) represent two frames of a video in the Desi class and the model recognized it as in Dot class since part of that firework video contains similar features as such in Dot class. In addition, even though Figure (e) and (f) are samples of Crosette class, both classified as belongs to Fish and WaterFlower classes respectively. Besides, Figure (g) recognized as in the Desi class even it originally contains in Dot class