Deep Learning for Computer Vision

Approaches for classifying videos

Videos have to be classified for several applications. Since the video is a lot of data, training and inference computations must also be accounted for. All video classification approaches are inspired by image classification algorithms. The standard architectures such as VGG, Inception, and so on are used for feature computation at a frame level and then processed further. Concepts such as CNN, attention, and LSTM learned in previous chapters will be useful here. Intuitively, the following approaches can be used for video classification:

• Extract the frames and use the models learned in Chapter 2, Image Classification, for classification on a frame basis.
• Extract the image features learned in Chapter 3