Video Analysis using Color spaces and RL
Jude Law · 2020
Color spaces have shown promising results in computer vision [1][2][3][4]. However, the use of color spaces in videos has not been analyzed. Video applications such as action recognition and video segmentation have been gaining a lot of attention in recent years. Using deep learning, action recognition results have reached a saturation point on some datasets [5][6][7]. In order to overcome these saturations, large scale datasets have been introduced such as the Sports1M [8] and Youtube8M[9]. Along with these datasets, the problem of limited data- namely zero-shot learning [10] has been introduced. ZSL has shown to be a very challenging problem because of introduction of unseen classes at test time. This is essentially a very practical problem and hence is of real value to the computer vision industry. To help optimize the learning with limited data, RL has shown great promise [11][12]. Along these lines, this work proposes the use of RL to determine what color spaces can help in video understanding namely action recognition and localization in the context of ZSL.