Mxnet provides very efficient and friendly APIs for DataIter for Images, which looks quite scalable. And it also contains components and networks for video learning, such as 3dconv and conv-lstm.
Is there any to-do/on-going project about extending the DataIter to a video reader, considering the up-coming trends of video recognition/understanding?
@HuichuanLiu that sounds like a great idea, I am not aware of anybody working on that right now. It would be nice to be able to process video files as easily as image ones. I would suggest to create a feature request in the github issue on the repo, so other people could register their interest under it and that way it can help the community to prioritize which features should be developed.