I want to create a 2 stream architecture for video classification using keras and tensorflow as its back-end .In this method you basically give 2 types of data to the model.One is the video itself(spatial information) and other being the optical flow images/files (temporal information) to the network as shown in the below link:

2 stream architecture in figure

My question is actually quite is easy.If you see the figure the optical flow data is fed into the model for training.My question is in what form is the data being fed into the model.Is it in the form of frames or an optical flow video generated from these frames(check the link to see how optical flow video and frames look like)? I have looked at multiple such type of papers and github repos for code implementation to find a solution but i have gotten into more confusion since they keep telling about .flo files in which the optical fields(temporal data) is stored.

Note:-Im just a beginner in this domain


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.