Video data represents each video as an ordered sequence of image frames. A dataset of videos is stored as a rank-5 tensor with shape (samples, frames, height, width, channels). section 2.2.8
Read the dimensions as follows:
- samples identifies the videos in the dataset. section 2.2.8
- frames preserves the sequence of images making up each video. section 2.2.8
- height and width describe each frame’s two-dimensional pixel grid. section 2.2.8
- channels store the values associated with each pixel. section 2.2.8
Representing a video dataset
A batch of videos can be organized as (samples, frames, height, width, channels): the first video is a sequence of frames, and each frame is an image represented by a height-by-width grid of channel values. section 2.2.8
