Is your feature request related to a problem? Please describe.
Richard's tutorial video_seg has a great application value on deploying segmentation models on video streams. However, trivial works like splitting the video to frames need to be done before inference if videos are given instead of images. Therefore, a video dataset accepting a video as input and frames of the video as output may be convenient to this use case and enable a fast video inference pipeline.
A video dataset like other datasets in monai but using a video as input and frames, ordered according to the frame index in the video, as output may be what users would like to be seen.

Describe the solution you'd like

• Implement a dataset with a video input and generate frames as output.